Home

Proceedings of the 5th Robotour Workshop

1. c ch soub n ve ty ech vrstv ch Senzorick vrstva zahrnuje moduly rozhran k d l m senzor m kter prov d n zko rov ov p edzpracov n dat nap odstran n chyby zp soben un e n m u gyr a data asov oraz tkuj Percep n vrstva transformuje m en do intern ch datov ch struktur re spektive vykon v vysoko rov ov zpracov n dat Prim rn mi moduly t to vrstvy jsou moduly pozi n estimace kter je dvoj ho typu lok ln a glob ln Mapovac vrstva se skl d ze dvou vnit n ch model sv ta jednak pro lok ln navigaci jednak pro glob ln navigaci Lok ln model je budov n mapov n m v sledk z modul detekce sj zdn ho povrchu dle polohy z skan z lok ln ho BA z 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 odhadu pozice do vnit n reprezentace zalo en na m ce obsazenosti Glob l n mapa nen autonomn tvo ena ale je do syst mu dod na z vn j ku ve for m tu OSM Kontroln vrstva se star o pl nov n a zen robota jak na lok ln rovni vyh b n se p ek k m udr en robota na cest detekce k i ovatek tak i na glob ln rovni rozhodov n na k i ovatk ch napl nov n trasy do c le Kognitivn syst m se dv ma vnit n mi modely sv ta dle m tka navigace p edstavuje innou architekturu e c celou adu st ejn ch probl m v au
2. ky byly patn identifikov ny i neidentifikov ny v bec p i couv n reagovaly ultrazvukov sn ma e pomalu tak e jsme do p ek ky narazili V leto n m ro n ku jsme skon ili na 2 m st co vzhledem k neslad nosti nov ho software bereme jako sp ch Z v r A koli ast v sout i nen prim rn m c lem projektu poskytuje n m velmi cenn reality check a motivaci dokon it funguj c e en v pevn m term nu D ky tomu se na e pr ce posouv rychleji 1 kdy t den p ed sout je velmi hektick co je ale asi p pad ka d ho t mu Se sou asnou architekturou syst mu a algoritmy jsme koncep n spokojeni nicm n je nutn dopracovat detaily Na e c le na nadch zej c obdob vylad n st vaj c verze software vy e en slab ch mist e dal vylep en algoritm e p prava na vytvo en roboauta na platform osobn ho vozidla Z v rem bychom r di pod kovali organiz tor m za v bornou organizaci sout e a v bec za realizaci tohoto zaj mav ho setk n nad enc v oblasti robotiky Sponzo i D kujeme za podporu na ich sponzor Artin spol s r o v voj software Sick spol s r o senzory a senzorick syst my 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Reference 1 S Thrun M Montemerlo and A Aron Probabilistic Terrain Analysis For High Speed Desert Driving Proc Robotics Science a
3. Convoy of robots HEAD BODY a TAIL is not supported Every robot must be able to carry 51 barrel of beer at least empty Exact position of destination will be known 10min before start Automatic start via timer Instead of PAUSE button only EMERGENCY STOP button is required It could be used in any dangerous situation for example in case of robot failure Pressing it means end of trial For the Big Red Switch is not possible to use a keyboard yes it is remarkable that still after several years some teams do not get it Just attach to your robot easily accessible red switch Without this BRS robot won t be accepted to the contest There is no restriction of an obstacle size There could be even cases when the whole path is blocked and it 15 impossible to avoid it There will be no extra beacon available There will be no category free style It is still possible to honk on an obstacle but it will have effect only in cases of curious grandma On the other hand bench won t notice honking The same holds for other robots or banana paper boxes Best Base Bratislava Daniel Zilinec Robert Najvirt Brmlab Prague Michal Tula ek Vaclav Hula CGS Robotics Italy Matteo Unetti Nicola Giordani Torguato Cecchini Eduro Prague Tom Roubi ek Jiri Isa Jan Roubi ek Istrobotics Bratislava Pavol Bosko Peter Bosko Odysseus Prague Jaroslav Halgasik Lenka Mudrova Matous Pokorny Petra K
4. e LEE We wanted to see a comparison of several approaches to the mobile robotics The competition gives us an opportunity to have our solution judged in an unbiased fashion R team After Istrobot and Eurobot I wanted to try something new e Roboauto The competition served as a motivation to finish a functional version of algorithms and of the robot We wanted to present our results to a general public We expected to meet with a like minded community e Radioklub P sek After seeing the robots in 2007 we believed we could do better 2 What did the competition give you e Propeler team We met people in the same domain of interest saw their approach and other technology Every year we have a motivation to catch up with our first result e LEE We have seen how a relatively simple solution by R team can solve a given task We realized that the increasing accuracy of hardware and sensors can have a huge impact on the accuracy of simultaneous localization and mapping We have been shown how important it is to deal with the technical details and with the reliability of the robots e R team I have learned that even the hardware is not fully reliable Indoor robots do not suffer from such prob lems I realized how difficult the task is even though I have expected some difficulties even beforehand e Roboauto It has fulfilled our
5. mo i8 12 Odysseus 0 2 4 H m 7 13 Bird 0 0 2 0 m 2 14 15 Istroboties 0 0 0 0 0 14 15 Smely Zajko 0 0 0 0 ys 0 Tatran Team Tren in and Radioklub P sek delivered full tanks and the total difference was only 2 meters under the measurements resolution So we decided to join these two places c 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 zs de 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robotour Solution as a Learned Behavior Based on Artificial Neural Networks Miroslav Nadhajsk and Pavel Petrovi Department of Applied Informatics Faculty of Mathematics Physics and Informatics Comenius University Mlynsk dolina 842 48 Bratislava Slovakia miroslav nadhajsky st fmph uniba sk ppetrovic acm org Abstract Our contribution describes a mobile robot platform that has been built for the purpose of the contest Robotour robotika cz outdoor delivery challenge The robot is a standard differential drive robot with a good guality consumer market digital video camera with a lightweight but high performance laptop computer used as the main control board Supplementary board is used to control motors and sensors of the robot The robot utilizes a behavior based architecture and its vision module that is responsible for track following is utilizing an artificial neural network that was trained on a set of imag
6. Sept 19 2010 Robotour robotika cz outdoor delivery challenge Ji Sa Faculty of Mathematics and Physics Charles University in Prague Malostransk n m st 25 118 00 Praha 1 Czech Republic Email isa ktiml mff cuni cz Abstract In this paper we present an international contest for autonomous robots Robotour robotika cz outdoor delivery challenge The main task is a navigation in real world situations First three years were held in park Stromovka Prague Czech Republic and raised an interest of many teams media and general public Last year the contest started to migrate To our knowledge there is no similar European outdoor contest for fully autonomous machines Note that there are some common fea tures with American Mini Grand Challenge and a younger Japanese Real World Robot Challenge The rules of Robotour are described in more detail together with experience gained over the past four years both from the organizers and the participants point of view Keywords autonomous robots outdoor international com petition I INTRODUCTION Competitions such as Eurobot 1 and DARPA Grand Chal lenge 2 have repeatedly shown that both young students and senior researchers are attracted by competitive research envi ronments Autonomous robotics is a multidisciplinary domain which offers educational opportunities and interesting real world research topics In 2004 the American Defense Advanced Resea
7. na se ty mi N FET tranzistory http www mlab cz Modules PowerSW NFET4X01A DOC NFET4X01A cs pd Jako nosn deska pro elektronick moduly byla pou ita z kladn deska 1621 MLAB http www mlab cz Modules Universal BASE162101A DOC BASE162101A cs pd Kontakty Martin Kakona martin kakona i cz Jakub Kakona kjakub gmail com Martin Povi er martin povik gmail com Kry tof Celba kristofc97 gmail com Roman Dvo k roman dvorak email cz Luka Cizek cizelu gmail com dd 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robozor esk Bud jovice Martin K kona Jakub K kona Martin Povi er Luk ek Josef Szylar Roman Dvo k Kry tof Celba http www robozor cz ROBOZOR cz Ant na ROBOTIC S CLUB robot Vector JADERN ELEKTR RNA TEMELIN HLAVN PARTNER Senzory 45 sth Robotour Workshop Bratislava Slovakia Sept 19 2010 Telemetrie E Pre o Q apices biz Sede E be DED Eiser a Urei Zi a e GO Ueleg heer EE TE Ma 5441 a Bae Mees sg EXLotesbHeselnont miE z biden erro TF Serieen m e ke Cat F m r nora Perez E rez uel DEA uss e EN EN rmn Pere 14 are U N k eien AEE Patt 5 e a maii veter o aa d ei Eet L Levy mowe came 1 artan ELO H nid ray ESKE E Brany meler rrikac ESAS runa vela can LL st HAT mp Ben i D bh t T n
8. spread from simple records of the traveled distance odometry and direction compass to a non trivial image analysis saving notable points along the way This technique of navigation restricts the robots only to places where their creator spent often several days of relatively tiring and exhausting work to build a very specific map It is relatively clear that to create a map for a robot to travel from Pisek to Opava won t be a one afternoon job with this technique But there is a map available to the general public where many people cooperate on its creation and this way the coverage is getting better every day This is possible only because everybody agreed on how the map should look like If ever a small group of people should create robot able to move in our world it is necessary that this robot uses our maps this means to use maps created by somebody else than the robot s author This is something we would like to endorse this year III FOR YEAR 2010 The fifth year of this contest should be a next step towards smarter and more autonomous robots In contrast to the previous years robots get only a map and coordinates of the destination The robots should be able to navigate around the park even 1f they have never been there before The map and the destination should be the only information the robots get before the start Robot successfully solving this task should be able to demonstrate its abil
9. tonomn ch mobiln ch syst mech pohybuj c ch se v rozs hl ch nestatick ch pro st ed ch v ce viz 2 4 Detekce sj zdn ho povrchu m kognitivn ho syst mu autonomn ho robota p i em robustnost p esnost a spolehlivost zvolen ch klasifika n ch metod implikuje sp nost cel platformy Bez dostate n rovn rozpozn n sc ny a jistoty rozpozn vac ch metod je del pohyb v dynamick m prost ed pro robota tak ka nemysliteln V detekci sj zdn ho povrchu a p ek ek se uplat uj data z laserov ho d lko m ru a barevn kamery Vlastn metoda detekce cesty je vybudovan na t chto p edpokladech cesta je tvo ena prostorov souvislou mno inou bod s v zna nou barvou cesta m n zkou m ru hrbolatosti a nulov nebo velmi mal sklon cesta je nejm n tak irok jako je ka robota cestu lze aproximovat rovinou D ky uveden m p edpoklad m lze definovat dostate n diskriminativn p znaky jako jsou hrbolatost prostorov souvislost barevn profil cesty aj kter umo uj relativn sp nou detekci sj zdn ho povrchu viz obr 2 5 Lok ln navigace Pod pojmem lok ln navigace se skr v problematika e en odhadu relativn pozice robota v i okoln m objekt m a zp sob interakce s bl zk m okol m jmenovit detekce sj zdn ho povrchu volba lok ln ho c le atp Jako abstraktn model prost ed pro f zi zpr
10. H ti P ne a p E 1 rr ia L L D a i D 1 u r A a D A T 4 D D D 1 4 PON 1 i j m D L qe 1 d Le p n i D H e m F 5 Fa i ZE a x Preco U u D E d D F F i 3 1 Sram kenge e Li Callana vulgaris Mires oby ajn ed m d z 8 a E wes mi ge EO ee 46 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Radioklub P sek Martin ern Pavel Hubka Karel Kozl k Milan ha Anton n Seiner Blanka Seinerov Martin Stejskal http www kufr cz RHE AAT TERME Bee PERE SDT PERE IEEJ Ho ow x wa rm I mna LU EE J 1 KS we me a LO Plvo a DREvo E e cun SE Ultrazvukov a infra erven nav d n 9 osy sn ma polohy MEAN GPS Navilog 303 Rizeni zaji uj dva mikroprocesory PC 800MHz ATMEGAS OS Debian Motory Valeo 12V s p evodovkou d c aplikace v jazyce Python Palubn nap t 12V z Rizeni pohan AVR ATMEGAS Rel ov H m stky m Rel ovy H m stek E Palubni nap ti 12V 41 7 EE WE Bratislava Slovakia Sept 19 2010 BOTANICK Z HRAD UNIVI IERE IT KRONE die SL I T JPU VEDECKA e ER 4 URLs Eu T g D JEA p Ca m SN LL AUTO be Zavod Je t na hotelu n m spadl a zasekl se ret z na LOko B hem 4 hodin se posunul as o 205 AM PM error 48 UNIVERZITA KO
11. Tsukuba Real World Robot Challenge RWRC took place in Japan for the first time 8 Since 2009 a similar straight line outdoor challenge takes place in P sek Czech Republic 9 Robotour robotika cz outdoor challenge 15 focused on au tonomous ground vehicles and their orientation in the real world outdoor environment The robots perform a delivery task in complex environments of city parks They are not allowed to leave paved roads Participants of various background are welcome In the previous years students from high schools university researches and hobbyists took part In this paper we describe the Robotour robotika cz outdoor delivery challenge General rules are covered in Sec tion II In Section III we share experience obtained from the organizers point of view Reflections of the par ticipants are captured in Section IV II RULES A Historical Overview The rules for each year change slightly and the contest becomes more and more challenging every year The unified theme of all years 1s robot s ability to autonomously navigate in outdoor environments and to move payload from one place to another The robots have to be fully autonomous which means that after a task entry they have to control themselves Since the first year the basic requirement is to navigate on paved roads in the park without leaving them similar to cars not leaving the streets In the second year a possibility of robot cooperation was
12. association robotika cz 15 an annual meeting of teams building and or programming outdoor robots that navigate in a city park filled with trails trees grass benches statues water ponds bridges and people The task changes every year but the main challenges are 1 be able to localize and navigate on a map supplied by the organizers and 2 be able to follow the trails and paths without colliding with the obstacles or leaving the path without reaching the goal See 1 for the exact rules of this year s contest Various solutions for the challenge were developed however in most cases they did not take advantage of advanced artificial intelligence algorithms In particular only few different vision algorithms were developed until today several teams shared the successful solution of 2 and many solutions rely on the use of odometry compass and GPS We would like to address this area and prepare a solution for the contest in 2010 or 2011 that will utilize AI algorithms The second author has participated in the competition team several times in the past and collected some experience and motivation for a new attempt In this article we describe the principles our solution is based on and is currently being built In the following sections we describe the mechanics and the hardware robot overall architecture the software components and the AI methods that we aim to use Finally we summarize the experience with building and progr
13. b t schopen demonstrovat sv schopnosti v jak mkoli parku s odpov daj c mapou Stejn jako v minul ch ro n c ch jsou podporov ni robustn j roboti schopn p ev et n klad Pro zv en div ck atraktivity bude start robot hromadn D le bude zavedena samostatn kategorie WagonOpen pro podpo en za naj c ch t m IV PRAVIDLA kol kolem robot je v zadan m asov m limitu 30min dopravit n klad do c le vzd len ho a Ikm Roboti mus byt pln samostatn nesj d t z cesty a spr vn se rozhodovat na k i ovatk ch podle zadan mapy M sto startu 1 m sto c le bude pro v echny roboty stejn Mapa Vektorov mapa chodn k v parku bude vych zet z vektorizace ortofotomapy a t my si ji mohou d le zp es ovat Z kladn idea je p evzata z Open Street Map Ve v sledku budou moci roboti pou t pouze tuto sd lenou mapu jak koli jin mapy jsou zak z ny Roboti T m m e letos nasadit pouze jednoho robota Ka d robot mus m t EMERGENCY STOP tla tko kter robota zastav Tla tko mus b t snadno p stupn erven a mus b t pevnou sou st robota aka Big Red Switch aby se v p pad hroz c ho nebezpe dalo snadno stisknout S robotem mus b t mo nost snadno manipulovat libovoln dv dosp l osoby ho mohou odn st n kolik des tek metr Je z rove definovan minim ln velikost na robotovi mus b t b hem cel sou
14. blokovan a p ek ku nen mo n objet Nebude k dispozici referen n maj k e Nebude kategorie voln j zda e Troubit na p ek ku je 1 nad le mo n ale efekt to bude m t pouze v p pad kdy je p ek kou nap klad zv dav babi ka Naopak lavi ka p es zvukovou podobnost obou zareagovat ciz robot nebo krabice od ban n 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robotour 2010 rules Martin Dlouhy a Zbyn k Winkler http robotika cz Abstract Fifth year of this autonomous outdoor robot contest will be held on 18th September 2010 It will take place in Slovakia in one of the preselected parks in Bratislava In contrast to previous years robots will get only map and coordinates of destination Note that robots will not know their exact starting position and interaction with operator will be reduced to entering destination Robot successfully solving this task should be able to demonstrate its navigation in any park with corresponding map I GOAL The objective of the Robotour contest is to encourage development of robots capable of transporting you to work in the morning or to deliver the building material you have just purchased in an online shop The path to this goal is neither easy nor short but we believe that the outcome is worth it Il Maps In the previous years there was an abundant mapping of the environment shortly before the contest itself These maps
15. debug the program it must be fully deterministic given the same inputs it must send exactly the same commands to the actuators of the robot For this reason there must not exist any non seeded random number generator in the code and only the appropriate logged CAN messages are used for timing information The logged GPS signal can be used for a global time coordination with the outer world The computer clock must not be used any where in the code or a new logged computer clock data source must be created If multiple threads are to be used a rigorous synchronization or logging of the interaction must be performed to guarantee the determinism These seemingly strict requirements implicate several cru cial properties of the code e If a possibly rare problem occurs during a test run we have the data ready to reinvoke it e When replaying from a log we can add testing statements without influencing the perceived timing e We can even stop the code in a debugger and inspect the variables without being afraid of any timeout kicking in and influencing the behavior In the default configuration when replaying data streams from logs we verify the determinism of the code If a com mand different from a logged one is produced an error occurs This works as a great test case for code refactoring We relax this condition when developing new code with a new behavior The real world data stream is fed into the new code to check
16. electric outlets B Experience of the Organizers There were couple lectures we have learnt over the last four years organizing Robotour and previously several years of organizing Czech Cup of Eurobot The basic scenario was already mentioned and serves good and is worth a recom mendation What has changed over the years are two major Accommodation is usually partially or fully sponsored Fig 2 Robot of the R team left leading the allied robot of RobSys right trends the number of teams is increasing and the task is getting more difficult In the first case we tried to find some optimal timetable of the rounds and we are still not satisfied What suits the teams does not suit a general audience and vice versa This year we will start all the robots from one place simultaneously which could be attractive for spectators but may cause problems to many teams The task complexity is another issue Beginners have a harder position to enter the contest every year For 2010 we discussed a new category WagonOpen but we will probably cancel it The reason is a new for the beginners with outdoor robots highly recommended contest Robotem rovn Robot go straight in Pisek In Pisek the task is to navigate as far as possible on a 3 meters wide and 300 meters long park road This is exactly the first stage which is necessary to enter the Robotour contest IV REFLECTIONS A Questions To reflect an influence the com
17. elektronika se skl d z n kolika modul MLAB jejich v et najdete n e a po ta e NetTopu Prestigio Jon NetTop Prestigio Ion disponuje dvoujadrovym procesorem Intel Atom taktovanym na frekvenci 1 6 GHz pam t RAM DDR2 o velikosti 2GB 2x1024MB a integrovanou grafickou kartou nVidia GeForce 9400 Pro ukl d n dat slou 2 5 pevn disk s kapacitou 250GB Po ta nav c disponuje esti USB konektory jedn m konektorem RJ 45 pro ethernet HDMI konektorem v stupem pro sluch tka a vstupem pro mikrofon K tomu aby se robot udr el na cest slou trojice teplotn ch senzor MLX90247 Navigaci v ter nu robotovi zaji uje GPS navigace a kompas CMPS03 Krom idel pro j zdu po cest jsme museli robota vybavit i ty mi ultrazvukov mi senzory SRF02 a dv ma ultrazvukov mi senzory MaxSonar EZ3 nebo je pot eba detekovat r zn p ek ky kter se na cest mohou objevit n hodn chodci lavi ky odpadkov ko e figurant zv dav babi ka atd D le jsme robotovi implementovali dva modifikovan senzory z inkoustov tisk rny HP pro m en rychlosti ot en h del obou motor a tak by p i dorovn n po tu ot ek m l robot jet rovn ale v praxi se uk zalo e rozd ln tlak vzduchu v pneumatik ch zp sobuje nezanedbatelnou v chylku kterou t mto zp sobem nelze kompenzovat Robot Vector dosahuje hmotnosti t m 40kg a rozm r 0 674m x 1 1m x 1 2m ka x d lka x
18. ho prost edn p ed vlastn sout Tyto mapy sahaly od jednoduch ch z znam ujet vzd lenosti odometrie a sm ru kompas a po netrivi ln anal zu obrazu z kamery a zapamatov n si v zna n ch bod Takov zp sob navigace ale p edur oval robot m pohyb pouze tam kde jeho tv rce str vil asto i n kolik dn relativn nam hav m a vy erp vaj c m vytv en m velmi specifick ch map Je pom rn z ejm e vytvo it t mto zp sobem mapu pro robota na cestu nap klad z P sku do Opavy nebude pr ce zrovna na jedno odpoledne Na druhou stranu ale existuj dostupn mapy na jejich tvorb se pod l mnoho lid a d ky tomu maj st le lep pokryt To je mo n jen proto e se v ichni shodli na tom jak m takov mapa vypadat Pokud m b t mal skupina lid n kdy schopn postavit robota schopn ho pohybu v na em sv t je t eba aby tento robot vyu val na e mapy to jest mapy kter je schopen vytvo it 1 n kdo jin ne autor robota A to jsme se rozhodli v tomto ro n ku podpo it III MOTIVACE PRO ROK 2010 P t ro n k by m l b t dal m krokem na cest k chyt ej m a autonomn j m robot m Na rozd l od p ede l ch ro n k roboti dostanou pouze mapu a sou adnice c le Roboti nebudou p esn zn t svoji startovn polohu a interakce s oper torem se omez na zad n c le Robot sp n e c tuto lohu by m l
19. http www elrob2006 org 2006 Real World Robot Challenge Outdoor robotic challenge http www ntf or jp challenge 2009 Robotem Rovn Outdoor robotic challenge http www kufr cz view php nazevclanku robotem rovne 2010 amp cisloclanku 2010020002 2010 Open Street Map Public http www openstreetmap org 2010 mapping activity 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robotour 2010 pravidla Martin Dlouhy a Zbyn k Winkler http robotika cz Abstrakt P t ro n k sout e autonomn ch outdoor robot se bude konat 18 z 2010 na Slovensku v jednom ze t p edvybran ch park v Bratislav Na rozd l od p ede l ch ro n k roboti dostanou pouze mapu a sou adnice c le Roboti nebudou p esn zn t svoji startovn polohu a interakce s oper torem se omez na zad n c le Robot sp n e c tuto lohu by m l b t schopen demonstrovat sv schopnosti v jak mkoli parku s odpov daj c mapou I C L C lem sout e Robotour je podpo it v voj robot schopn ch dopravit v s t eba r no do pr ce nebo v m p iv zt stavebn materi l co jste si pr v objednali v online obchod Cesta k takov mu c li nebude ani jednoduch ani kr tk ale v me e v sledek bude stat za to Il Mapy V p edchoz ch ro n c ch bylo hojn vyu v no mapov n sout n
20. introduced In the third year obsta cles were added and robots had to deal with them successfully In the fourth year robots did not know exactly their start position and had to deal with obstacles more carefully The fifth year of this contest should be a next step towards smarter and more autonomous robots In contrast to the previ ous years the robots get only a map and coordinates of the des tination The robots should be able to navigate around the park even if they have never been there before The map and the destination should be the only information the robots get before the start Robot successfully solving this task should be able to demonstrate its ability with a corresponding map in any park 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Fig 1 in 2009 A simple map of the Lu anky park in Brno given to the participants B Detailed Rules 1 Task The task for the robots is to deliver payload in a given limit of 30 minutes to a destination as far as 1 km Robots must be fully autonomous not leave a road and choose correct path on junctions The starting place starting time and the destination will be the same for all the robots 2 Map Vector map of footpaths in a park will be based on a vectorization of an ortophotomap and teams could im prove it further The basic idea 15 taken from Open Street Map 10 A robot is allowed to use only this shared map all other maps are prohibited 3 Robots
21. organizaci sout e a v bec za realizaci tohoto zaj mav ho setk n nad enc v oblasti robotiky Tato pr ce vznikla ste n za podpory grantu FIT VUT v Brn FIT S 10 1 Bezpe n spolehliv a adaptivn po ta ov syst my a v zkumn ho z m ru V zkum informa n ch technologi z hlediska bezpe nosti MSM0021630528 Reference 1 S J Julier and J K Uhlmann A new extension of the Kalman filter to non linear systems In Proc of AeroSense The 11th International Symposium on Aerospace Defence Sensing Simulation and Controls 1997 2 D Herman Lok ln navigace autonomn ho robota diplomov pr ce Brno FIT VUT v Brn 2010 E07 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 E 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Dokumentace robota Vector Luk ek cizelu gmail com 24 srpna 2010 Robot Vector je outdoorovy robot vyvijeny od roku 2007 leny klubu Robozor Jakub Kakona Martin Povi er Kry tof Celba Roman Dvorak Lukas i ek pod vedenim Ing Martina K kony Byl vyv jen pro sout Robotour ale v roce 2010 odstartoval i na sout i Robotem rovn v P sku Robot je postaven na ty kolov m podvozku z invalidn ho voz ku s n honem p edn ch kol kter jsou poh n na elektrick mi motory s p evodovkou nap jen mi z 12V olov n ho akumul toru kter zaji uje nap jen v ech elektronick ch sou st robota d c
22. prijima pripojeny cez bluetooth na strane PC cez virtu lny COM port x Spracovanie obrazu detekcia preva ne zeleno zafarben ch objektov tr va rastliny listy vyh banie sa oblasti mimo cesty x Pou itie Dijkstrovho algoritmu pre ur enie najkrat ej cesty vo vektorovom grafe yalozenom na importe OSM exportu mapy v chodzej polohy a cie ov ho bodu am x Platforma NET jazyk C x Pouzit kniznica pie video eegend obrazu snimaneho web kamerou AForge NE x Pouzitie Dijkstrovho Ee pre ur enie cesty z grafu zostaven ho z d t OSM exportu Start polohy a ciela x Vyhodnocovanie stavov snima ov z prednej asti robota detekcia prek ok x Komunikacia COM rozhranie cez USB konvertor radiaca elektronika Bluetooth virtualny COM GPS USB web kamera THE END 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 E List of Authors D Dlouh Martin 1 Drah nsky M 33 H Herman David 33 I LSAS e RO E d Gaius aki 1 21 N Nadhajsk Miroslav 15 Najv rek Jan 33 O Ondr ek Tom 33 Be poesia dut ise err 33 P Petrovi Pavel 15 R Roub ek Tom A Roub ek Jan A Cizek NIA 39
23. the interesting difficult problems first suggesting that we can make it work in the end Our team focuses on the other path We keep com bining reliable blocks and adding reliable improvements over the time This path has two key advantages over the attractive one 1 It works If we run out of time at any moment we have a working solution which can compete 2 It is more fun Because the robot moves from the very beginning it provides a motivation to add improvements reliable working path reliability of a solution unreliable attractive path smart smartness of a solution dumb Fig 1 There are two basic paths from an unreliable dumb solution lower left to a reliable smart solution upper right 1 The attractive path suggest focusing on the complex stuff first and making it work later 1 The working path advises to keep improving reliable solutions and see their immediate impact This is much more rewarding than digging five tunnels from five directions and praying they meet in the end and in time III TEAM Our team is a mixture of hardware and software experts who have been involved in robotics for about a decade Jan Roub ek designed the winning platform for Cleaning Contest 2002 Lausanne and constructed several more successful plat forms Explorer Eduro Eduro Maxi for several more competitions Eurobot RobotChallenge Field Robot Event later Tom Roub ek focuses on ele
24. v ka Komunikace mezi robotem a program tory je zaji t na pomoc WiFi Access Pointu D le je k robotovi p ipojen upraven joystick s p ti tla tky Na robotovi je nainstalov n Linux s j drem verze 2 6 31 16 Konkr tn se jedn o distribuci Ubuntu 10 04 Lucid Lynx P i programov n robota bylo pou ito n kolik programovac ch jazyk C C Lula a Python 39 sth Robotour Workshop Bratislava Slovakia Sept 19 2010 Nyn n sleduje n kolik fotografi kde jsou vyzna en d le it sou sti robota EN WiFi Router Nosn desky s d c elektronikou eb A l A e e k i TS em w RC gt AN E 40 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 iP Sd GX RTT EIE d Konvertor RS232 na TTL urovne RS2328INGLE01A A USB232R01B D 22 na RS232 TTL Pohled na d c elektroniku a senzory F RS ZE 1 1 d I x k E SA lt x k GP A Ai eg b 7 be AA x RA v i Ad by f 2 L 3 2 j j i y e e I d t n 3 A d k 1 F EA u a a N k l A a ZE Z ZA AE Ge T A j e e g L k 44016 Mde m eis a 4 m E TE bk a L E d y d A x L edita Pe d 100 Konvert 1573 E A ow l P bk 11 9454 1 Hi e ra i sla w amp amp 8 i w P e da
25. za vedouc m robotem Ka d robot vag nek bude m t rampu na um st n maj k pro napojen v ce robot v kolonu Detaily budou up esn ny v FAO1 Omezen rozm r a bodov n za inteligenci a za n klad bude p evzat z hlavn sout e tak aby hardware robota mohl b t vyu it v dal m ro n ku v hlavn kategorii V ODLI NOSTI OPROTI MINUL MU RO N KU e Hromadn start v ech robot z jednoho p edem nezn m ho m sta Je zak z no mapovat si sout n prostor p ed vlastn sout M sto toho mohou t my vyu vat mapu na jej tvorb se mohou aktivn pod let Kolona robot HEAD BODY a TAIL ji nen nijak podporov na Ka d robot mus b t schopen uv st 51 sud piva alespo pr zdn e P esn poloha c le bude zn m 10min p ed startem Automatick start pomoc asova e e M sto PAUSE tla tka bude pouze vy adov no EMERGENCY STOP tla tko kter m e kdokoliv pou t v p pad nebezpe n situace nap klad poru e robota Jeho stisk znamen ukon en pokusu Za Big Red Switch nen mo n pova ovat kl vesnici ano je pozoruhodn e 1 po t ch letech to po d n kter m t m m nen jasn Prost dejte na sv ho robota snadno dostupn erven vyp na Bez odpov daj c ho BRS nebude robot do sout e p ipu t n Neni omezen na velikost p ek ky Mohou tedy nastat situace kdy bude cel cesta
26. 83 P DS LOM 15 295 DJ I IC 15 20 15 Er H m LET ZZ TZE LOM i5 GOI Culkwir Ho EUR E LAT Ee path ete mach P m Aki He Com fd CHE ust m 9 Delta Trgt DEO Recompuhe during replay Licker om 6 77 Bgrume GE during remegeg Beta Caibeate angie ee keon 7727Z7Z TT i i I Se Sepeni SEMSORPAD GORRIZ LTZKO save wis nes SO GEk 1 E EN 2010 17122000 12 3 2020 aen Fast Hlavn obrazovka SensorPadu 28 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 VideoMap Modul pro zpracov n video vstupu z kamery um st n na sto ru vozidla pro lep rozhled J assi EC EL Ei r oo Ta B Wess Ei VideoMap Rozpoznan cesta v parku 29 sth Robotour Workshop Bratislava Slovakia Sept 19 2010 RoboMap Aplikace zaji uje zpracov n vstupn ch dat d ky nim modeluje mapu okol vozidla a pl nuje budouc trasu Aktu ln verze dok e rozpoznat vozovku dle mapy sj zdnosti a tvaru identifikuje k i ovatky v etn v jezd a rozpozn v statick a dynamick p ek ky Dap map Los boka map randa ad Lana DRMLETLOM SO TZON ENER Eer Gm CA ease icu Y Ei eo FH E LI u ES Tots G Saz DEIERRI Eze SA o Er ma L balia naj mem ali ET LO ENTE pur er OF m
27. A team can deploy multiple robots this year but only a single designated one is used to compute a score Every robot must have an emergency stop button which stops its motion The button must be easily accessible red and must form a fixed part of the robot aka Big Red Switch so it could be used in a case of a danger The team must show that it 15 easy to manipulate with the robot two people must be able to carry it several tens of meters There is also a minimal size robot has to carry 51 beer barrel at least an empty one 4 Leaving the Road The robots are expected to stay on the road which means to stay on the paved passage ways If any robot leaves the road its trial ends The team has to take care of their robot and remove it immediately 5 Obstacles There could be obstacles on the road Besides natural obstacles like benches there could also be artificial ob stacles A typical artificial obstacle is for example a figurant a banana paper box or another robot Robots must not touch an obstacle Contact with an obstacle means an end of a trial The robot may stop in front of an obstacle and visually or acoustically give a notice Note that the robot has to detect that the obstacle is no longer present 6 Robots Interaction Situations in which a faster robot catches up with a slower one will not be explicitly handled The faster robot can handle the slower robot as an obstacle i e avoid it or wait until the o
28. MENSK HO BRATISLAVA V HOTELOVEJ ASTI 027 601 992 18 OQ E PN OKT BRA V CASE OD AV DO 187 HOD gt wa OD 9 DO 18 HOD a 3 V SOBOTU A NEDELU 9 DO 18 HOD S VYSOKO KOLSK INTERNAT DRU BA PON KA UBYTOVANIE KONGRESOV PRIESTORY RELAX CENTRUM 09 2 288 000 ON 654 2004 WERTEZFITFTFTEWITWITT NN flan A podruh 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Pod kov n ale nakonec jsme t et a tvrt Jak na to aneb co ne d lejte Ned lejte zm ny strategie Na nikoho se nespol hejte Hlavn se sna te ned lat zm ny t sn p ed startem Zkontrolujte si as Dot hn t roubky alespo 2x do roka Testovat testovat testovat Organiz tor m za vynalo enou energii a usili E V em len m na eho t mu Robotem rovn 2011 49 s V ichni jste srde n zv ni ji na 3 ro n k sout e Robotem rovn z Kdy 21 5 2011 z Kde V P sku Palack ho sady we We 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 an 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Brmlab Praha Michal Tul ek V clav Hula http brmlab cz Tescoma spherical mirror DrmiaD hackerspace prague Brmbot Outdoor cy DC MIabD hackerspace prague http brmlab cz The place for your ideas and projects Lots of hackerspaces in Europe Germ
29. Proceedings of the 5 Robotour Workshop held within the Robotics in Education conference Editor Richard Balogh Sept 19 2010 Bratislava Slovakia Published by Robotika SK oz http robotika sk robotour Credits Cover design David Jablonovsky Cover photo Richard Balogh IXIFX editor Richard Balogh using IATEX s confproc class September 2010 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 CONTENTS About the Contest Martin Dlouh Jiri Sa Robotour robotika cz outdoor delivery challenge Robotour 2010 Pravidla Robotour 2010 Rules Robotour 2010 Team List Robotour 2010 Results White papers Pavel Petrovic Miroslav Nadhajsky Smely Zajko Team RoboTour solution as a learned behavior based on Artificial Neural Networks Jiri Isa Tom Roub ek Jan Roub ek Eduro Team Tom Ondr ek Jan Najv rek Pavel Brzobohat Pavel ernock David Herman Ji Zbirovsk Vojt ch Robotka Roboauto Ouido D Herman T Ondr ek J Najv rek F Ors g M Drahansk Roboauto Team Robot pro Robotour 2010 Luk ek Robozor Team Robot Vector Presentations Robozor esk Bud jovice Radioklub P sek Brmlab Praha Tatran Team Tren n List of Authors 111 11 13 15 21 25 33 39 45 47 51 53 55 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 iv 5 Robotour Workshop Bratislava Slovakia
30. Workshop Bratislava Slovakia Sept 19 2010 Score The team whose robot manage best to proceed along the route wins Critical is the aerial distance of the last robot position leaving the road a collision or a timeout to destination For every meter towards the destination team gets one point If the team carries payload it is awarded twice points for the payload Each robot can carry only one payload Payload is 5liter beer barrel full In every round a robot can get at most twice the aerial distance of the start and the destination Organization The contest will consist of 4 trials for each team The start and destination will be different for every trial The selected destination will be announced to all teams 10 minutes before the start The speed of the robots is not important actually it is limited to 2 5m s All points gained during all trials will be summed together The trial starts at a specified time and ends after 30 minutes The robot must leave the start area within 10 minutes of the start If the robot does not move for a 60 second its trial ends Each team has to arrange for one person familiar with the rules that will be part of the referee team during the competition Homologation A team can participate in the contest if it is able to score at least one point Necessary condition is the ability to travel 10 meters long route fragment without collision with an obstacle The start procedure will be tested automat
31. a sebou p i jedn stran cesty V r mci startovn oblasti m e t m um stit robota podle vlastn ho uv en Po ad robot na startu bude dan v sledky v p ede l m kole lep robot bude bl e k c li V prvn m kole bude pozice ur ena po ad m sp n homologace Roboti startuj automaticky pomoc vnit n ch asova Minutu p ed startem u nesm doch zet k dn interakci s robotem gt 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Bodov n Vyhr v t m jeho robot bude trasu nejl pe zdol vat Rozhoduj c je vzdu n vzd lenost posledn pozice vyjet z cesty kolize i vypr en asov ho limitu k c li T m z sk v 1 bod tzv bod za cestu za ka d metr sm rem k c li vzd lenost start c l vzd lenost kone n pozice c l Za vezen n kladu z sk v t m dvojn sobek body za n klad Ka d robot m e v zt jeden n klad N kladem se rozum 51 pivn soudek pln V dan m kole tedy robot m e z skat nejv e po et bod roven dvojn sobku vzdu n vzd lenosti start c l Pokud robot neopust startovn oblast z sk 0 bod Organizace Sout bude m t 4 kola Pro ka d kolo bude vybr n jin start a c l Vybran c l bude ozn men 10 minut p ed startem kola Rychlost v t to sout i nehraje roli je omezena na 2 5m s Do celkov ho v sledku se s taj body za v echna kola Kolo za n v
32. a virtual serial port over radio BlueTooth connection In debugging and testing applications the robot can be controlled using a wireless gamepad connected using a proprietary 2 4GHz radio link In general the robot is designed in such a way that it can be used in many different applications For instance a stereo vision system or an arm with a gripper can be installed in the cargo hold area Additional sensors can be easily mounted on the aluminum profiles or wooden base Fig 3 shows overall system architecture 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 analog video camera frame grabber serial port USB converter BlueTooth radio or wired serial port 9 DOF IMU compass accellerometer gyroscope ultrasonic Fc SBot Board 1 bit serial port distance with AVR odometry sensors analog microcontroller encoders position controllers interrupt Figure 3 System hardware architecture IV SOFTWARE CONTROLLER ARCHITECTURE The software architecture is tailored for the Robotour contest In this years contest the goal for the robot 15 to navigate to the target without knowing its starting location It is only given the target coordinates and an official map of the park It may not use other map information The software controller is logically divided into five main components see Fig 4 The first component planning uses the map with the destination location and genera
33. acovan ch senzorick ch dat byla zvolena 2D m ka obsazenosti kde hodnoty bun k reprezentuj m ru jistoty rozpozn vac ch al goritm cesty v rohodnostn p stup Kl ov m prvkem senzorick mapy je pevn sou adn soustava ur en po te n m bodem ke kter mu jsou lokali zov na v echna m en Mapov n v stupu detekce prob h dle dat z lok ln ho pozi n ho estim toru obr 1 naviga n architektura Lok ln pozi n estim tor je de facto virtu ln m senzorem agreguj c m data z odometrie kompasu a inerci ln jednotky na z klad dynamick ho modelu rob ota s vyu t m UKF Unscented Kalman Filter 1 Data z GPS senzoru nejsou 35 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Traversable segments x m o o A o o T T T l l l I Roughness groups 1 L 1 1 1 1 l l 1 5 4 3 2 1 0 1 2 3 4 5 y m Obr zek 2 Detekce cesty pomoc lidaru detekce cesty pomoc kamery d ky sv povaze chyb agregov na kalmanov m filtrem ji v tomto bod nebo pro p esnost lok ln mapy p i lok ln navigaci v jednotk ch metr nen rozho duj c a ani podstatn akumula n chyba ale plynul zm na mezi ud van mi polohami vozidla pro zaji t n korektn ho mapov n v sledk detekce Samotn pl nov n pohybu robota je postaveno na sad p edem definovan ch man vr je odpov daj fyzick m mo nostem platfo
34. al hundred meters in our conditions Robotour 2006 and 2007 Aiming for a good competitive behavior not a winning one we see that our desired traveled distance is comparable with a possible GPS noise On the other hand odometry allows us tu travel such a distance safely For these reasons and because of our general approach of iterative improvements we decided to use odometry for the global localization For navigation we developed a simple application which allows us to enter waypoints manually We would prefer an autonomous planning instead of the manual mode but this is the point where the competition deadline stopped our efforts The basic submodules are combined in a straightforward manner Global localization together with the predefined tra jectory give us a direction of the current segment Given ZD 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 the direction information from the odometry with the compass we know our directional deviation Next the road recognition algorithm gives us another hint where to go We merge these two using a weighted average The resulting desired direction is put into the obstacle avoidance algorithm which finds the most appropriate collision free direction VI FUTURE IMPROVEMENTS A Hardware During the competition we observed our robot to be con fused on steep surfaces The 2D compass is influenced by its tilt and our robot entered a junction in a very unfortunate di
35. amming the robot up to date H MECHANICS The robot is a simple robot with differential drive kinematics with one supporting free rolling caster wheel The length of the sides of its square base is 45 cm the air inflated wheels of a diameter 15 3 cm are mounted on the outside of the base in the front of the robot The total weight is about 6 kg without any load The robot provides a storage space of ca 20 x 20 x 45 cm to carry a heavy load approx 5 kg which can be placed close to the center of rotation above the propelled wheels so that it does not have a negative impact on maneuverability of the robot The main control unit is a portable computer mounted in a flat plastic frame with a foam to compensate the shocks The lead acid 12V 9Ah rechargeable battery being the heaviest component is stored under the base between the wheels keeping the centre of gravity low Color camera with a true optical image stabilizer and CCD image sensor is mounted using anti shock foam on a U shape construction frame built of aluminum profiles together with GPS and IMU sensor see Fig l The camera is inclined 10 downwards The IMU sensor must be mounted far from any sources of electric and magnetic fields such as motors and wires Placing GPS high compensates also for obstacles in the surrounding terrain which may hinder the GPS satellites signal The robot is built from raw materials except of the motors wheels and consoles that hold them
36. anel with buttons and indicators are attached to the base These modules are made from aluminium pro files and sheets and have many mounting points for simple extending Eduro robots use SMAC Stepper Motor Adaptive Control drives This is an original Robsys technology for gearless drives which is based on closed loop controlled stepper motors Eduro Maxi has the motors attached by a simple belt transmission The brain of the robot is a single board computer running Linux OS The computer is equipped with AMD Geode CPU running at 500 MHz with 256 MB RAM compact flash card wi fi 3 Ethernet 1 RS232 and 2 USB ports RS232 port is dedicated for CAN bus connection via transparent RS232 CAN bridge High data throughput without data loss is secured by real time serial driver Eduro uses CAN bus as its main communication network All sensors and actuators with low data rate requirements are connected through the CAN CANopen is the preferred communication protocol Camera and laser range finder are connected directly to the main computer via Ethernet Except for CAN and Ethernet DC and 1 wire Dallas buses are used in the robot I2C is not designed for large distances I2C inter integrated circuits therefore it is used for short local buses only The 1 wire 1s used for a diagnostic network and advanced power management Distributed power switches thermometers battery chips and other simple sensors and modules are connected via th
37. anetarium Praha in the first park Stromovka did A good idea is also a combination with an exhibition of robots and a related technology parallel to the contest There is no registration fee but the teams have to take care of catering and pay an accommodation Small items remain on the bill leaflets printing diplomas cup for the winners and a Saturday night dinner The dinner is usually sponsored and the goal is to unite the teams and give them a chance to relax a little bit after the contest Note that prices are rather symbolic which lowers expenses on one side and also reduces a potential rivalry between the teams A Duties over the Year The first task of the organizers is a precise specifi cation of rules for the next contest They are presented on the robotika cz website in Czech and English languages The core remains the same autonomously navigate in a park and the changes are usually a consequence of a discussion at the workshop and experience gained The second task is to ensure an affordable accommodation for a relatively large group of people 50 people needed accommodation in 2009 An agreement with a university dormitory serves well The reservation must be performed usually a month in advance and that defines a clear deadline for the registration of the teams Finally it is necessary to find an interesting park manage permission for the contest day and find building with large enough room s for team base with many
38. any Iceland Hungary http hackerspaces org Bratislave Progressbar http www progressbar sk Brno Underground hackerspace OK1KPI P sek Jain us 51 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 D 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Tatran Team Tren in Michal Kuku ka Juraj E ery Marek Sutliak http www tatran team szm com PECIFIKACIA ROBOTA V EOBECN INFO Platforma postaven na p sovom podvozku x Implementovan ru n dialkov ovladanie Vlastn koncepcia mechanika riadiaca elektronika softv r spolahlivost odolnost Jazdn vlastnosti v ter ne pou it cenovo dostupnej ie umelohmotn p sy r chle opotrebovanie R x 2 x jednosmern 24V motor s prevodovkou x Riadiaca elektronika regul cia pomocou PWM elektronika spracovania stavov sn ma ov x Komunik cia riadacej elektroniky pomocou RS232 pou ity vlastny protokol Al 8958252 program nap san v assembler vstup r chlosti a smery ot ania motorov v stup aktu lny stav snima ov x 5 x opticky dif zny sn ma Balluff detekcia prek ok pred robotom x 2 x IR snima snimanie bo n ch prek ok pri ot an neimplementovan x USB WEB kamera 800 x 600 pixels detekcia cesty x GPS Navilock BT 413 detekcia aktu lnej polohy 54 e LOKALIZACIA x GPS
39. aplanova Propeler Opava Tomas Kotula Martin Kotula Eva Kotulova Roboauto Brno Jan Najvarek Tomas Ondracek Pavel Brzobohaty Vojt ch Robotka Roboauto Quido Brno David Herman Jirka Zbirovsky Sirael Praha Kamil Rezac Jaroslav Sladek 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robotour 2010 Team List Robozor Cesk Bud jovice Martin Kakona Jakub Kakona Martin Poviser Luk ek Josef Szylar Roman Dvorak Krystof Celba Radioklub Pisek Martin Cerny Pavel Hubka Karel Kozlik Milan Riha Antonin Seiner Blanka Seinerova Martin Stejskal Short Circuits Praha Pavel Jiroutek Dan Polak Lukas Polak Smely Zajko Bratislava Pavel Petrovic Miroslav Nadhajsky Tatran Team Trencin Michal Kukucka Juraj Ecery Marek Sutliak URPI Team Bratislava Marian K ik Michal Bachraty Organizers Martin Dlouhy Zbynek Winkler David Obdrzalek Ondrej Luks Frantisek Duchon Richard Balogh 11 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 E 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robotour 2010 results Order Tem Ban 1 Run2 Run 3 Run4 Full Total 2 Roboauto Quido 84 0 53 16 no 153 6 Short Cireuits Praha 17 2 43 9 mo 7 7 Rebzo 0 0 3M ys 70 8 CGSRoboties 0 2 27 0 ys 58 9 Proper 20 2 5 19 mo 4 1 bik 9 0 O0 9
40. asnosti pou ita pro m en n klon a jako gyrokompas Odometer na p evodovce p ipojen do notebooku p es RS232 Magnetick kompas p ipojen do notebooku p es I2C Kamera pro lok ln orientaci a ur en pozice p ipojen p es FireWire kr tko dosahov ch UZ sn ma pro bezpe n couv n p ipojen p es I2C Softwarov e en Samotn softwarov e en je koncipov no formou nez visl ch aplikac kter spolu spolupracuj p es protokol TCPIP 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 TCP IP f E VideoMap zpracov n videa LIDAR Odometr z sk n dat a TCP IP Komnas eu i Rizeni I i volantu a I l I motoru l TCP IP Architektura aplikaci Z kladn mi softwarov mi moduly jsou SensorPad zaji uje p ipojen a z sk n dat ze v ech sn ma Tyto data pak poskytuje pomoc TCPIP streamu dal m aplikac m Ka d aplikace si m e vybrat kter data chce ze SensorPadu z sk vat Aplikace vytvo ena v Delphi RoboDriver zaji uje spojen s v konovou elektronikou vozidla p ij m po adavky na zen a ty aplikuje Dok e p ij mat v ce zdroj zen a podle priority je aplikovat nap Collision Avoidance modul zabra uj c sr ce m vy prioritu ne samotn inteligentn zen Aplikace vytvo ena v D
41. aviga n architektura UKF lidar ka mera 1 vod V posledn ch letech je v civiln sf e popul rn v zkum autonomn ch mobiln ch vozidel slou c ch pro transport osob nebo zbo 5 problematikou v ech po hybuj c ch se syst m zce souvis ot zka jejich navigace je zast e uje velice slo itou sekvenci pod loh po naje sb rem dat ze senzor lokalizac tvorbou modelu okoln ho prost ed pl nov n m a kon e samotnou realizac ustanoven ho pl nu Nadto je t eba aby robot um l e it ne ekan situace kolize s p ek kami apod nebo re ln prost ed v n m se pohybuje p edstavuje vysoce dynam ick syst m 2 Experiment ln platforma Jako z klad pro vytvo en pohybliv ho vozidla s pracovn m n zvem Karl k je pou ito modifikovan d tsk voz tko P gPerego z tvrzen ho plastu Pro bezpe n uchycen senzor byl p id n kovov r m z oh ban ho plechu kter je pevn spo jen s kostrou robota K r mu je p ipevn na p ihr dka pro uchycen notebook Kv li p em st n t i t byly p esunuty baterie z p edn sti auta do sti zadn D le bylo nahrazeno h ebenov zen servomotorem kter je p mo p ipojen na t hla kol Snaha o potla en ot es vedla k v m n origin ln ch plastov ch kol za kola gumov a tak k odpru en podvozku viz obr 1 mechanick konstrukce robota 33 5 Robotour Workshop Bra
42. bstacle disappears In general the road rules will be respected right of way avoidance to the right passing on the left 7 Start All robots will start from the same park road simultaneously A minimum width of this road is 3 me ters The starting area for each team will measure approx 1 5 x 1 5 meters Starting areas will follow one after another on one side of the road Within the starting area each team can place its robot as they see fit The order of the robots on the start is given by their results in the previous round a better robot will be closer to the destination The order in the first round will be given by the order of successful homologation Robots start automatically via their internal timers During the last minute before the start no interaction with the robot is allowed 6 Score The team whose robot manages to proceed along the route best wins The aerial distance of the last position of the robot leaving the road a collision or a time out to the destination is critical For every meter towards the destination a team gets one point If the team carries a payload its score is doubled points for the payload Each robot can carry only one payload A 51 beer barrel full serves as a payload In every round a robot can obtain points at most equal to twice the aerial distance of the start and the destination 9 Organization The contest will consist of four trials for each team The start and dest
43. ce lok ln mapa 7 Z v r Navigace autonomn ch mobiln ch robot p edstavuje komplexn lohu je se koncentruje do schopnosti p esunu robota z aktu ln pozice na pozici c lovou Pouze b h robota v re ln m prost ed m e zaru it spr vnost a pou itelnost uveden ch metod proto e samotn po ta ov simulace nen schopna postihnout v echny aspekty kter mohou ve skute n m okol robota nastat A proto a koliv ast v sout i nen prim rn m c lem projektu Roboauto poskytuje n m velmi cenn reality check a motivaci dokon it navr en e en v pevn m term nu V leto n m ro n ku sout e byl nast n n p stup pou it nez visle na dvou odli n ch robotech aby se ov ila obecnost cel ho n vrhu Mimo vlastn sout byl ka d robot schopen bezchybn absolvovat des tky metr dlouhou dr hu Bohu el komplikac se stal hromadn start kdy byl v t robot Karl k prakticky vy azen ze hry kv li nedostatku sn ma pro detekci velmi bl zk ch a mal ch objekt Nicm n men robot Auido se um stil na p kn m druh m m st V dal m roce se zam me p edev m na v voj protikolizn ho syst mu lep pl nov n pohybu robota a odlad n n zko rov ov ho zen nebo pr v tyto sti se uk zaly v tomto ro n ku sout e jako kritick mi a nejm n odlad n mi Pod kov n R di bychom pod kovali po adatel m za v bornou
44. ctronics and low level software CANopen firmware for sensors and actuators Ji I a keeps an eye on artificial intelligence computer vision probabilistic reasoning etc We also build on an extensive work by Martin Dlouh His long experience in robotic systems reflects itself in the framework he developed for Eduro and Eduro Maxi for RobotChallenge Field Robot Event and few other robotic competitions in 2010 Sadly he cannot be a team member for Robotour IV HARDWARE A Platform Eduro Maxi 15 a modular mid size mobile robotic platform designed as both a teaching tool for higher education and a re search platform for academia and industry 2 The platform 15 designed around a tricycle base with two differentially driven wheels and one caster wheel Fig 2 The on board electronics P 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Fig 2 Eduro Maxi Laser scaner IP camera Sick LMS100 Motor R CAN R5232 Main computer converter x86 Linux System Display managment beeper k a 1 Tracking unit GPS GSM EE ls Sprayer Fig 3 Hardware structure consists of smart sensors and actuators connected by a CAN bus The main controller module is implemented as a single board x86 based computer running Linux OS The base of the robot is a construction module that includes a battery and motors Modules such as the caster wheel or the control p
45. d Robotour robotika cz outdoor delivery challenge its rules and their evolution over the time We share experience gained while organizing several years of the com petition and show several patterns worth following The com petition has been successful in attracting people to robotics and giving them an opportunity to learn The contestants enjoy a chance to meet others exchange ideas and compare their approaches in an independent manner As the competitors note while seemingly simple the competition became difficult to participate in This in turn led to a creation of two new robotic competitions in Czech Republic which differ in the level of difficulty Currently there exists an evolutionary path for a person interested in robotics through these outdoor competitions up to Robotour and possibly even further 1 2 3 4 5 6 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 REFERENCES Eurobot Contest for amateur robotics http eurobot org 1997 DARPA Grand Challenge http www darpa mil grandchallenge 2004 David Obdr lek Eurobot 2010 Robot Revue May 2010 Robot Challenge Indoor robotic competitions http robotchallenge org 2010 Istrobot Contest for amateur robotics http robotika sk 2010 Field Robot Event Agricultural robotic competition http fieldrobotevent de 2010 7 8 9 10 ELROB Ist European Land Robot Trial
46. dy v ur en as a kon po 30 minut ch Robot mus opustit startovn oblast nejpozd ji do 10 minut od startu Pokud se robot mimo startovn oblast nebude 60 sekund pohybovat bude aktu ln pokus ukon en Ka d t m mus zajistit jednu osobu znalou pravidel kter bude b hem sout n ho dne pat it do t mu rozhod ch Homologace T m se m e z astnit sout e pokud uk e e je schopen z skat alespo jeden bod Nutnou podm nkou je projet desetimetrovy sek bez kontaktu s p ek kou Testov na bude startovac procedura automatick start a funk nost EMERGENCY STOP tla tka Pou it tekutin ravin pyrotechnick ch materi l a iv ch bytost je zak z no Ka d robot bude b hem j zd doprov zen jednou osobou z t mu star 18 let kter je za jeho chov n zcela zodpov dn Technick dokumentace Ka d t m dod ke sv mu robotu robot m z kladn technickou dokumentaci pro prezentace ve ejnost a novin e V t zn t my 1 a 3 m sto pak budou po d ny o podrobn j dokumentaci pro webovou prezentaci a tedy zjednodu en zapojen nov k do sout e v n sleduj c m roce Kategorie Vag nkyOpen roboti BODY a TAIL z minul ch ro n k se mohou z astnit nov kategorie WagonOpen Jej hodnocen bude nez visl na hlavn sout i a hlavn motivac je umo nit zapojen nov k Tato sout bude m t 3 kola a c lem bude se udr et
47. e 1 wire bus 26 and 1 wire bus are connected to the CANopen network through gateways The power supply is provided by two 12 V 8 Ah sealed lead acid batteries The whole robot uses a single power source to simplify management The motors are powered from 24 V supply branch directly from batteries The main computer CAN network and most of sensors are powered from a stabilized 12 V branch The auxiliary 5 V power supply is present for simple connection of the low cost sensors B Sensors The Eduro carries many sensors but only few of them were really used in Robotour contest The collision avoidance system employs laser range finder the road detection algo rithm uses data from camera and global navigation was based on compass and odometry The GPS sonar and other robot sensors were not used The data for odometry is acquired from drives The drives send actual extended shaft angles every 50 ms with resolution of approximately 0 2 mm The relative position and orientation of the robot is calculated from this data The absolute robot orientation is determined by a compass The compass is a part of the inertial unit Currently we use the two axis compass HMC6352 from Honeywell It is a one chip solution with I2C bus however the chip is not visible from CAN The data from compass and other sensors are periodically polled by the CAN module processed and then forwarded to the central unit The azimuth readings from the sensor are converted
48. elphi Videomap zpracov n videa identifikace cesty a p ek ek vyu v 1 informac z Lidaru Vytvo eno v jazyce C s vyu it m knihovny OpenCV LidarMap pou v informac z Lidaru pro rozpozn n cesty a p ek ek Vytvo eno v jazyce Delphi RoboMap zaji uje zpracov n vstupn ch dat d ky nim modeluje mapu okol vozidla a pl nuje budouc trasu Pomoc propojen na RoboDriver d samotn vozidlo Aplikace vytvo ena v jazyce Java a P sth Robotour Workshop Bratislava Slovakia Sept 19 2010 SensorPad Aplikace zaji uje p ipojen a z sk n dat ze v ech sn ma P ehledn zobrazuje stav spojen se sn ma i a tak slou k vizualizaci z kladn ch m en ch veli in nap Lidar Umo uje 1 ukl d n v ech dat a jejich p ehr n pozd ji vhodn pro diagnostiku a simulov n testovac ch p pad a senur Pad Raboauta CI Gecke te wes r mes Cergera Poste 12 r11 Fi 62570 m D Ds 412990m 3 54 i Mis Deeg QzTmi 051 a Carver stiermen 25 8 20 LATE 8 227817 DAO LZ js VP I PA LG Li k Ak GOGO re ii LE Rol EN yag Pa LISILISILBIRS Miihe 5I L5 65 74 LzJ L25 LB X9 am MEG DNUS Ari ste p updmeTme 122X1 Stret eg ZE VK DET BME Yan ET ve as E LE Ju as B PS Fix i i GET cs Ge db 6 Dike Speed C 27 ms Gear DUI 49 223 5 0
49. er barrels in 2010 from a starting point to a specified destination On their journey through a city park the robots must not leave paved roads Points are awarded for a distance traveled towards the destination point Before the competition the robots are given a map Open Street Map which they can use for navigation The robots need to perceive their environment e g where the road is and where the obstacles are to deal with the task successfully Our team Eduro Team won this competition in 2010 In this article we explain our approach Sec II and introduce the team members Sec III who developed the hardware Sec IV and the software Sec V which we describe too II APPROACH Reliability beats smartness any time The only better thing is reliability with smartness on top Smartness with an added reliability does not exist Over the last years we have seen over and over again that reliable partial solutions outperform incomplete advanced solutions Simply put The robot which moves is better than the one which does not This is especially important for us amateur roboticians with an extremely restricted time budget Fig 1 depicts two distinct paths we can take In the beginning we do not have any robot we start in the lower left corner of the diagram with an extremely dumb and unreliable solution none We want to reach the upper right corner with a reliable smart solution The attractive path seduces us to
50. er the space of possible headings that can be projected to the input frame where the headings leading to more path areas are more likely than those leading to less path area Input from the odometry and gyroscopes helps this module to improve its estimation of the path using its previous estimations and the relative displacement of the robot The obstacle recognition module 1s responsible for detecting obstacles in the planned path of the robot and for stopping the robot in case of a possible collision early enough so that avoidance could be attempted by the coordination module The robot is currently equipped with three ultrasonic distance sensors front ahead front left front right and thus the module reports on its output whether the path is blocked completely or only partially and also what is the size of the expected free buffer in front of the robot sensors ath recognition bstacl iti using map P g obstacle recognition gt plan Figure 4 Overall controller architecture ultrasonic distance The most complex module 15 the coordination module Its purpose 1s to take the prioritized outputs from the other three modules and to determine the best possible angular and linear velocity for the next instant movement When the confidence of the module is getting low the robot slows down If the confidence falls even lower the robot stops and starts rotating left or right depending which direction is expected
51. es This is a novel solution that has not been used in Robotour contest previously and our early experiments demonstrate promising results Keywords robotour navigation artificial neural networks learning robots I INTRODUCTION Applications of robotics technology in both production and personal use are becoming possible with the development of new materials motors sensors and vision ever decreasing cost of computing and memory capacity and development of new algorithms and control strategies Robots must be able to operate in dynamic and unpredictable environments Therefore one of the most important challenges to be solved reliably is robot navigation in both indoor and outdoor environments The robots must be able to localize themselves on a supplied map create their own map representations of the explored environment and they must be able to navigate their environments safely without colliding with obstacles or failing to follow the paths roads trails and tracks The real improvements in the technology typically occur when there is a large motivational pressure to produce a working solution This might either be a goal to produce a final product or alternately with somewhat more relaxed reguirements and settings which are suitable for experimentation and research when the goal is to develop a robot to participate in a robotics contest Robotour robotika cz outdoor delivery challenge organized by the Czech
52. expectation The competition gave us a practical experience with deploying a robot We have got an inspiration for further improvements of the hardware and algorithms We feel in touch with people with similar interests e Radioklub Pisek We realized the competition 15 not as simple as it seemed for the first look and few others 3 What were you disappointed with e Propeler team We are not really disappointed When the robot works everything is fine Answering the question What does the robot do is difficult when the task difficulty 1s not obvious e LEE Although there is a lot written by the competitors at robotika cz every year someone new comes and repeats previous mistakes R team In my opinion the competition has become too difficult Only one or two best teams can fully cope with the rules e Roboauto Problems with a reliability and with a robustness are bigger than we have expected We are disappointed with only a small media atten tion We hoped to get an attention of potential sponsors or future team members which has not happened so far e Radioklub P sek We are sad that the cooperation of multiple robots 1s not encouraged any more We have learned several interesting things doing that On the other hand as the competition evolves it does not suffice to copy a solution from the previous year V SUMMARY We have introduce
53. for coding errors and strange conditions but the new commands are not compared to the old ones B Robotour module The code for Robotour consists of four parts e Obstacle avoidance e Road recognition e Global localization e Navigation The main requirement of the 2010 rules is the obstacle avoidance During the start of each round there are many robots in a confined area Some of the robots try to move other robots are broken and form fixed obstacles Before we even start to think about global navigation and road following we need to get out of the starting area For this reason we implemented a modified VFH algorithm 3 As an input of the algorithm we do not use a local map but a current laser scan After the robot gets out of the starting area we expect to have several meters or tens of meters to the first junction To keep the robot on a road we combine the VFH obstacle avoidance typically there are objects such as shrubbery along the road with a vision based road recognition a her itage from our Short Circuits and Cogito MART history We also tried to detect a roughness of a surface with the laser scanner but we did not succeed Three key observations from past competitions formed our opinion about global localization for this year 1 A well kicked ball could rank third or second in 2009 2 Under the trees GPS error of 30 m is no exception 3 Purely odometric robots with a compass can travel sever
54. ic start as well as the functionality of the EMERGENCY STOP Usage of liguids corrosive pyrotechnic material and live beings 15 strictly prohibited Every robot has to be accompanied by a team member older 18 years who is fully responsible for the robot behavior Technical documentation Every team has to provide basic technical documentation about their robot for presentations general public and journalists Three winning teams will be asked for more detail description for website presentation and easier entry of novices in the next year Category WagonOpen Robots BODY a TAIL from previous years can participate in a new category WagonOpen Scoring of this category will be independent of the main contest and the primary motivation 15 to give a chance to novice teams This contest will have 3 rounds and the goal is to follow a leading robot Every robot wagon will have a ramp for placing beacons to link together several robot in a convoy Details will be specified in FAQI Robot restrictions and score for intelligence and payload will be adopted from main contest so that the robot hardware could be used for the next year participation in the main category 10 V BASIC DIFFERENCES AGAINST THE PREVIOUS YEAR Simultaneous start of all robot from one beforehand unknown place It is prohibited to map contest area before contest itself Teams can instead use shared map on which creation they can actively participate
55. ination will be different for every trial The selected destination will be announced to all teams 10 minutes before the start The speed of the robots is not important actually it is limited to 2 5 m s All points gained during all trials will be summed together The trial starts at a specified time and ends after 30 minutes The robot must leave the starting area within 10 minutes of the start If the robot does not move for 60 seconds its trial ends Each team has to arrange for one person familiar with the rules that will be part of the referee team during the competition 10 Homologation A team can participate in the contest only if it is able to score at least one point Another nec essary condition is an ability to travel along a 10 meters long route fragment without a collision with any obstacle The starting procedure will be tested the automatic start as well as the functionality of the emergency stop button Usage of liquids corrosive or pyrotechnic material as well as live beings is strictly prohibited Every robot has to be accompanied by a team member older than 18 years who is fully responsible for the behavior of the robot 11 Technical Documentation Every team has to provide basic technical documentation about their robot for presen tations general public and journalists Three winning teams will be asked for a more detailed description for a website presentation and to make the entry of novices in the following yea
56. into 1 100th of degree and sent over CAN bus as a 16bit integer The update rate is 20 Hz The HMC6352 is only a two axis magnetometer therefore tilt compensation is not possible The inertial unit including the compass is mounted on top of the pole away from sources of magnetic fields and ferro magnetic objects The module itself is covered by a plastic case and no steel parts are used During experiments we observed substantial changes in sensor readings caused by even seem ingly minor attachments to the pole such as a small umbrella therefore a caution is needed A presence of ferromagnetic objects can be compensated by the system but that requires recalibration and usage of non linear transformation The robot vision uses data from CCTV IP camera with fish eye lens The camera offers resolution of 1280x1024 pixels with frame rate 30 fps but due to limited processing power the quarter resolution 640x512 jpeg images with much lower frame rate are taken The last but very important sensor on Eduro Maxi robot is the laser measurement system SICK LMS100 We used it only for collision avoidance system in Robotour contest but its applicability is much wider It is a relatively expensive sensor but provides very useful data The LMS100 scans robot neighbourhood in a plain and gets distances to the nearest 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Road recognition Python framework Fig 4 Software architectu
57. into 100 rectangular regions of equal sizes that covered the whole image Each region formed an input to a neural network and the whole region was about to be classified as path or not path However the resulting resolution of the classified image was not satisfactory even after a further reduction of the region size so that the image was divided into 2500 segments Therefore we decided to use a sliding region For almost every pixel in the image we define a corresponding region it s larger neighborhood which forms the input vector The classification output produced by the network for each pixel in the image is then a real number from O to 1 estimating how much the network believes the pixel lies on the path Two examples of images that were not used in the training phase are shown in the Fig 6 b x E Figure 6 Examples of path recognition We used the RPROP training algorithm for multilayer perceptron in particular the implementation that is present in the OpenCV package The training used tens to hundreds of manually classified images from various places in a park with ite various path surfaces light and shadow conditions Since this is still an ongoing work and only preliminary results are available we restrain from a statistical analysis of the results at this moment and refer the reader to the page dedicated to the project with detailed results and data 5 Once the network is trained and
58. ity with a corresponding map in any park Robust robots able to carry a payload are preferred as in previous years The start will be concurrent from one place to increase spectators attractiveness There will be also new category WagonOpen to support novice teams IV RULES Task The task for the robots is to deliver payload in given 30 minutes time limit to destination as far as Ikm Robots must be fully autonomous not leave the road and choose correct path on junctions The place of start and destination will be the same for all robots Map Vector map of footpath in park will be based on vectorization of ortophotomap and teams could further improve it The basic idea is taken from Open Street Map At the end robot could use only this shared map all other maps are prohibited Robots Team can deploy only one robot this year Every robot must have EMERGENCY STOP button which stops its motion The button must be easy accessible red and must be fixed part of the robot Big Red Switch so it could be used in case of danger The team must show that it is easy to manipulate with the robot two people must be able to carry it several tens of meters There is also minimal size robot has to carry 51 beer barrel at least an empty one Leaving the road The robots are expected to stay on the road which means to stay on the paved passage ways If any robot leaves the road the trial ends The team has to take care of their robo
59. le and better robots in the next years REFERENCES 1 J I a and M Dlouh Robotour Robotika cz Outdoor Delivery Chal lenge in Proceedings of Robotics in Education RIE 2010 Bratislava Slovakia September 2010 pp 89 93 2 M Dlouh J Roub ek and T Roub ek EDURO Mobile Robotic Platform for Education in Proceedings of Robotics in Education RIE 2010 Bratislava Slovakia September 2010 pp 107 111 3 I Ulrich and J Borenstein WFH Reliable Obstacle Avoidance for Fast Mobile Robots 1998 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Projekt automatick ho zen vozidla ROBOAUTO ROBOCAR Model Ouido Robotour 2010 lenov t mu Tom Ondr ek Jan Najv rek Pavel Brzobohat Pavel ernock David Herman Ji Zbirovsk Vojt ch Robotka Ob ansk sdru en Roboauto tym roboauto cz Abstrakt Tento l nek popisuje architekturu modelu robotnick ho vozidla Auido vytvo en ho v r mci projektu Roboauto a pou it ho na sout i Robotour 2010 Z kladem modelu je model sk podvozek kter je osazeno t mito sn ma i LIDAR odometr GPS IMU inerci ln jednotka elektronick kompas 25 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Uvod V r mci projektu Roboauto bylo v 2 f zi vytvo eno modelov voz tko kter dostate n simuluje chov n re ln ho vozidla a je schopno n st sn ma e a za
60. mponents of the software platform can be reused in other applications the robot can be extended with stereo vision or manipulator We have designed implemented and tested in this context a new method for path recognition which is based on artificial neural network that is trained on a set of static images that are similar to the environment where the robot is to be operating We are currently working on integrating all the components of our prototype so that it could perform in its first 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robotour contest this year In the remaining 10 months of the project we will analyze the results from our participation and propose implement and verify improvements so that the robot can serve both as a competitive platform in the contest and as an educational tool in the course Algorithms for AI Robotics which is provided at our department to students of Applied Informatics ACKNOWLEDGMENT This work has been done with the support of the grant Podpora kvality vzdel vania na vysok ch kol ch from Nadacia Tatra banky REFERENCES 1 Robotour robotika cz outdoor delivery challenge Rules online http robotika cz competitions robotour cs last accessed August Ist 2010 2 K Ko nar T Krajn k and L P eu il Visual Topological Mapping in European Robotics Symposium 2008 Heidelberg Springer 2008 pp 333 342 3 G Bradsky A Kaehler Learnning OpenCV Co
61. mputer Vision with the OpenCV Library O Reilly Media Inc 2008 4 D Gu tafik SBot v2 0 Educational Robot for Clubs and Classrooms User manual to Sbot robot 2008 online http webcvs robotika sk cgi bin cvsweb checkout robotika sbot 2 0 doc dokumentacia_en pdf 5 M Nadhajsky Robotour diploma project page 2010 online http virtuallab kar elf stuba sk mnadhajsky N 0 0 i l 0 45 40 22 0 45 5 36 31 5 27 225 18 135 9 45 Angle in degrees n o e gt 0 O e E 5 x a 0 D D he o gt 4 ii 315 36 405 45 9 135 18 225 27 Figure 8 Two scenes after path recognition The bars show the average pixel intensity of pixels inside of triangles for a range of different rotations for both of the resulting images blue dark for the left image red bright for the right image 19 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 _ 20 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Eduro Team Ji I a Tom Roub ek Jan Roub ek Abstract Eduro Team won Robotour Robotika cz Outdoor Delivery Challenge 2010 To make the life of future participants easier and to satisfy a curiosity of current participants we describe our hardware software and the whole approach in this article I INTRODUCTION In Robotour Robotika cz Outdoor delivery Challenge 1 fully autonomous robots deliver objects 51 be
62. nd Systems Philadelphia PA USA August pages 16 19 2006 2 D Stavens G Hoffmann and S Thrun Online Speed Adaptation using Supervised Learning for High Speed Off Road Autonomous Driving In Proceedings of the International Joint Conference on Artificial Intelligence 2007 IJCAIO7 Hyderabad India 3 Ob ansk sdru en Robotika Sborn k Robotour 2007 sborn k ze sout e ve Stromovce r 2007 4 D Herman Lok ln navigace autonomn ho mobiln ho robota Diplomov pr ce VUT v Brn Fakulta Informa n ch Technologi Ustav Inteligentn ch Syst m Brno 2010 331 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Robot pro Robotour 2010 D Herman T Ondra ek J Najvarek J Zbirovsky P Brzobohaty V Robotka F Orsag a M Drahansk Vysok u en technick v Brn Fakulta informa n ch technologi Bo et chova 2 612 66 Brno esk republika iherman orsag drahan fit vutbr cz tym roboauto cz http www fit vutbr cz http www roboauto cz Abstrakt Tento lanek popisuje architekturu robotick ho syst mu pro Robotour 2010 kter byl vytvo en pod z titou projektu Roboauto Z kladem modelu je podvozek automobilov ho typu jen nese celou adu senzor pro detekci cesty a lokalizaci robota v parkov m prost ed V tomto ro n ku sout e byli nasazeni dva velikostn rozd ln roboti aby se ov ila obecnost vytvo en ho konceptu Keywords robot robotour n
63. o KT E lt E gt Ves e d d Joa 95 Ee s I SE l i cS a Pohled na d c elektroniku a senzory 41 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Vonn budi e b s i H L Pohled na d c elektroniku a v konov H m stky 42 51 Robotour Workshop Bratislava Slovakia Sept 19 2010 h del K Motory Pohled na senzory snimajici po et ot ek h del motor 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Na robotovi byly pouzity tyto moduly MLAB HB1FETO1A Vykonovy H mustek http www mlab cz Server GenIndex GenIndex php path Modules H Bridgeklang cs RS232SINGLEO1A P evodn k RS 232 na TTL rovn http www mlab cz Modules CommSerial RS232SINGLEO1A DOC RS232SINGLEO1A cs pd USB232R01B P evodn k USB na RS 232 s TTL rovn mi http www mlab cz Modules CommSerial USB232R01B DOC USB232R01B cs pd PIC16F84S01801A Modul pro procesor PIC http www mlab cz Modules PIC PIC16F84DIL1801A DOC PIC16F84DIL1801A cs pd PIC16F87xDIL2801A Modul pro procesor PIC http www mlab cz Modules PIC PIC16F87xDIL2801A DOC PIC16F87xDIL28014 cs pd PIC16F87xTQ4401B Modul pro procesor PIC http www mlab cz Modules PIC PIC16F87xT04401B DOC PIC16F87xT04401B cs pd BATPOWEROZA Modul sp nan ho 5V zdroje http www mlab cz Modules PowerSupply BATPOWERO2A DOC BATPOWERO2A pd NFET4X01A Modul v konov ch sp
64. os mas a F E KS Pea 1 ES Tots G Lis retary mi Ezez AAA EBE r kens EGE EAK HETA T 11 77 p pe pr ra rien PAT Tr phoq Ken Eed Farmaka KM SEis BIS annus EF d Uri al I RoboMap identifikace kfi ovatky 30 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Pr b h sout e Pro tento rok jsme zcela p epracovali dic software Robomap kter nov rozpozn val vozovku podle tvaru v map sj zdnosti rozpozn val k i ovatky a v jezdy z nich v lok ln map a ty pak mapoval na glob ln OSF mapu um l napl novat trasu j zdy pomoc man vr kter je vozidlo schopno prov st a to v etn couvac ho man vru v p pad Ze rozpoznal odbo ku p li pozd Sout e Robotour jsme se z astnili ji pot et pou v me tedy nabyt ch zku enost z p edchoz ch ro n k k tomu abychom neopakovali klasick chyby neotestovan zm ny t sn p ed sout spol h n na GPS i magnetick kompas Nicm n b n mu probl mu nedostatku asu na p pravu nov verze se ubr nit neum me Ten n m v leto n m ro n ku zp sobil e automatick zen vozidla nebylo odlad no tak dob e jak by mohlo Proto jsme b hem v ech kol m li n sleduj c pot e obj d n p ek ky nebylo p esn nar eli jsme do obj d n p ek ky z boku senzor 17 p ek ku nevid l n kter p ek
65. petition has had on its participants we have asked some of the past successful teams few questions 1 What did you expect from the competition 2 What did the competition give you 3 What were you disappointed with B Asked Teams The following teams were asked e Propeler team Opava A group of high school students who placed 2nd in 2006 e LEE Prague Researchers and students from Czech Tech nical University in Prague Winners of the year 2008 and the year 2009 e R team Rychnov nad Kn nou A team of a high school teacher Since 2010 he organizes RobotOrienteering in Rychnov nad Kn nou R team finished 2nd in 2008 in a coalition with the RobSys team see Figure 2 e Roboauto Brno A self funded group of researchers which ranked 2nd in 2009 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 e Radioklub P sek P sek Hobbyists and professionals who also teach electronics in a club Radioklub P sek got a 3rd place in 2009 Since 2009 the club organizes Robotem rovn mentioned in Section III C Answers 1 What did you expect from the competition e Propeler team The competition motivated us to build our first robot Having almost no restriction on the dimensions of the robot allowed for a simple construction We could use a notebook get an image from a camera and use a bought chip to control the motor and the servo we did not understand microchips and servos at that time
66. produces the classifications for the image frame pixels the path recognition module enters a second phase when it tries to evaluate all possible travel directions headings with respect to the chances that the robot will stay on the path For this purpose the module analyzes a family of triangles of the same area with the base at the bottom of the frame and the third vertex placed in the middle of the image For each such triangle we compute an average path likelihood The triangle for which the path is most likely i e where most pixels lay on the path is likely to be the correct new heading However the module outputs a full distribution over all possible headings so that the coordination module can take advantage of this information for instance to determine different directions at a heading or when trying to resolve ambiguous cases Fig 7 depicts the analyzed family of triangles Two example pictures are further analyzed in Fig 8 where the bars show how likely it 1s that following in the various directions is a good idea in order for the robot not to leave the path Figure 7 Triangles representing different turning projected to the image of recognized path VI CONCLUSIONS AND FUTURE WORK We have designed and implemented a robotic hardware and software platform to be used in the Robotour contest for outdoor robots navigating in park environment The hardware platform is implemented in a general way and most co
67. rch Projects Agency DARPA organized the first Grand Challenge The goal of DARPA was to foster a research in fully autonomous vehicles In the first year only 11 78 km of the 240 km long route were completed by the best team Al ready in the second year of the competition 2005 five vehi cles finished the 212 km long route This shows a tremendous impact the challenge has had on the field of fully autonomous ground vehicles Since 1994 the Eurobot competition attracts many young people more than 2000 in year 2010 3 Eurobot has suc cessfully shown how an international competition can be used to teach young people how to cooperate and how to develop complex systems In 2006 the Robotour robotika cz outdoor delivery chal lenge has been founded In our opinion the large gap in com plexity between Eurobot like competitions e g RobotChal lenge 4 Istrobot 5 and other and competitions like DARPA Grand Challenge needed to be bridged In about the same time other organizers felt similar insufficiency and more com petitions were born Since 2003 Field Robot Event focuses on the agricultural automation 6 Since 2006 European Land Robotic Trial allows research teams and industrial companies Martin Dlouhy MapFactor s r o tef nikova 24 150 00 Praha 5 Czech Republic Email md robotika cz to demonstrate their unmanned outdoor systems in realistic scenarios and terrains 7 One year after Robotour in 2007
68. re when connected to the real hardware left and when reading data from logs right Python framework objects The field of view is 270 degrees with a resolution of 0 5 or 0 25 degree The detectable distance is up to 20 m which is sufficient for most small mobile robotic applica tions The data is provided by a simple TCP IP interface with an update frequency of 25 Hz The LMS100 is appointed for indoor use but it is usable in outdoor environment too Only a strong direct sunlight can make troubles The sensor is reliable enough therefore we do not use any other sensor for obstacle detection V SOFTWARE Our high level code is a combination of C C and Python modules C is used for parts with strict timing requirements e g real time communication over the serial line Vision based road recognition is programmed in C The biggest part of code is formed by our self made Python framework The specialized Python code for Robotour is just a thin layer on top of the aforementioned modules Fig 4 A Logging Every single data source CAN network camera LIDAR and GPS is logged into a separate log file Further the frame work keeps another log to record the order of arrival of the data frames from different sources This allows us to i check process and debug each data source independently on the other sources 11 replay the whole information stream for debug ging of the whole code To make it possible to
69. rection which in turn forced the robot to leave by a wrong road Because Eduro Maxi is a grown up small Eduro it could be reasonable to change the motors for stronger ones This year we also enjoyed a very nice weather during the competition but we cannot rely on it for the next year We need to make the robot more water proof We are also running on the performance limits of the on board PC B Software Our incremental development stopped at the point of navi gation and localization For the next year we wish to e tune the odometry possibly with the 3D compass e develop a more complex localization possibly incorpo rating the inaccurate GPS Ade e build and use a local map for obstacle avoidance e run an autonomous trajectory planner Without the local map it is possible for our robot to leave the road while avoiding an obstacle With the map the standard VFH can be used to prevent this issue During the competition after the robot turned wrong the absence of a planner prevented it from recovering and going back VII SUMMARY In this article we have described our focus on reliability and incremental improvements Sec II and how it was reflected in our Sec III hardware Sec IV and software Sec V We also reveal current weaknesses of the robot and how we want to handle them Sec VI We hope this helps current and future teams to organize their work and choose such solutions that there are more reliab
70. rmy a na algoritmu A Volba lok ln ho c le souvis s estimovanou pozic robota na glob ln map Pakli e se robot nach z v bl zkosti k i ovatky a byla li detekovan tato k i ovatka i v lok ln map tak v b r lok ln ho c le reflektuje po adovan odbo en vozidla spr vn m sm rem V opa n m p pad se lok ln c l vol na st ed cesty ve vhodn vzd lenosti od robota 6 Glob ln navigace Kombinace glob ln ho polohovac ho syst mu a intern ho naviga n ho syst mu INS m dopl uj c se vlastnosti GPS poskytuje glob ln pozi n informaci s n zkou frekvenc av ak bez integra n chyby Pr v n zkou frekvenci a ob asnou nedostupnost dat lze kompenzovat INS jen nen z visl na vn j ch sign lech a p edstavuje spojit m c syst m Navr en lokaliza n metoda pro glob ln lokalizaci slu uje pozi n informaci z v ce zdroj pomoc fuzzy logiky a vyu v metrick ch a topologick ch in formac obsa en ch v map Glob ln odhad pozice robota je nejprve z sk n z GPS senzoru N sledn je provedena konfrontace v sledku s dodanou mapou ve form tu OSM aplikov n m porovn vac techniky point to line a p padn ko rekce glob ln pozice obr 3 Pro nalezen cesty z po te n pozice do c lov byl pou it Dijkstr v algoritmus 6 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Obr zek 3 Metoda glob ln lokaliza
71. rs easier III ORGANIZATION Robotour is organized as a three day event Friday to Sun day Friday is dedicated to the testing clarification of rule details and homologation During the homologation we want to make sure that robots are not dangerous have a functional 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 emergency stop button and are able to gain at least one point in the contest Saturday is the contest day Finally there is a workshop on Sunday It is after the contest so the competitors have a fresh experience with their robots and algorithms They are also not stressed any more and thus this is a good moment for sharing knowledge We started to enforce this three day template after the first competition in 2006 That competition ended on Saturday and most teams left without letting us and other teams know what has worked and what has not What was even more important was that teams left exhausted from the programming marathon and one team had a car accident on the way home Since the following year the workshop is mandatory The Robotour contest is relatively self supporting and the expenses are minimal There is no special playground a public park is used instead There is no need for renting a hall because the event takes place outside To be precise some room is necessary as a base for the teams especially in bad weather conditions It is recommended to have a partner who provides this place like Pl
72. t e um st n 51 pivn soudek alespo pr zdn Vyjet z cesty Je dovoleno se pohybovat pouze po parkov ch cesti k ch Pokud robot sjede z cesty aktu ln pokus pro n j kon O jeho v asn odklizen se mus postarat sout c t m P ek ky Na trase se mohou nach zet p ek ky Krom p ek ek p irozen ch lavi ky atp mohou b t na tra um s ov ny 1 p ek ky um l Za typickou um lou p ek ku se pova uje nap klad figurant pap rov krabice od ban n i jin robot Roboti nesm vej t v kontakt s p ek kou Kontakt s p ek kou znamen ukon en pokusu Robot m e p ed p ek kou zastavit a vizu ln i zvukov upozornit e p ek ka byla detekov na Fakt e p ek ka u nen p tomn mus roboti detekovat sami Interakce robot Situace kdy rychlej robot dojede robota pomalej ho nebude nijak zvl tn e ena Rychlej robot se m e k pomalej mu zachovat nap klad jako k p ek ce tj objet ho nebo po kat a odjede s m Obecn budou respektov na pravidla silni n ho provozu p ednost zprava vyh b n se vpravo p edj d n vlevo Start V ichni roboti budou startovat sou asn na jedn z parkov ch cest v ichni stejn Minim ln ka cesty na kter se bude startovat je 3 metry Startovn oblast pro jeden t m bude m t velikost cca 1 5x1 5 metru Startovn oblasti budou um st ny t sn z
73. t and remove it immediately Obstacles There could be obstacles on the road Besides natural obstacles like benches there could be also artificial obstacles A typical artificial obstacle is for example a figurant a banana paper box or other robot Robots may not touch an obstacle Contact with an obstacle means end of the trial The robot may stop in front of the obstacle and visually or acoustically give notice Note that the robot has to detect that the obstacle 15 no longer present Robots Interaction In the cases where a faster robot catches up a slower one won t be explicitly handled The faster robot can handle the slower robot as an obstacle 1 9 avoid it or wait until the obstacle disappears In general the road rules will be respected right of way avoidance to the right passing on the left Start All robots will start simultaneously from the same park road A minimum width this road is 3 meters Start area for each team will have approx 1 5x1 5 meters Start areas will follow one after another on one side of the road In the start area each team can place its robot as they see fit The order of robots on start is given by the results the in previous round a better robot will be closer to the destination The order in the first round will be given by the order of successful homologation Robots start automatically via their internal timers A minute before the start is allowed no interaction with the robot 5 Robotour
74. tes a path plan for the robot to follow It tries to minimize the number and complexity of the crossings as these are the most critical places and candidates for navigational errors The component outputs a seguence of locations that are to be visited by the robot Whenever reguested the module can generate a new plan after a problematic place in the map has been reached The second component localization using map 15 responsible for the most accurate localization of the robot on the map It is using the information from the compensated compass IMU for heading from GPS for position estimation and from the position encoders to estimate the distance traveled and turns made All the information 15 integrated and with the help of the map and the path plan the target distribution is determined using a probabilistic Monte Carlo estimation The output of the localization module 1s a probabilistic distribution over the expected heading in the very next correct movement and the expected distance to the next crossing or target The third module path recognition is the most important one for the actual control of the motors and has a priority over the localization module It receives the image from the front camera and recognizes which parts of the image correspond to the path and which of them correspond to other surfaces The next section explains this procedure in more details The output of this module is again a probabilistic distribution ov
75. that includes several axes of gyroscopes accelerometers and magnetometers thus compensating for various robot inclinations when traveling uphill or downhill This is important since the simple compass sensors provide incorrect information once the robot and thus also the sensor 1s tilted Finally for the visual input we chose to use a standard video camera Panasonic SDR T50 due to a very good ratio of parameters price The video camera is built around a CCD sensor which has the advantage over the CMOS image sensors of taking the image instantly Cheap CMOS cameras therefore suffer from a serious vertical distortion when the camera is moving since the different rows of the image are scanned at different times In addition the camera has a built in true optical image stabilizer which further compensates for distortions due to the movement Unfortunately we found this stabilizer to be insufficient and thus we have supported it with an anti shock foam placed between the camera and the platform where it is tightened using flexible textile tape The camera renders its image either as 16 9 or 4 3 image however it sends a wider signal down to its video output jack connector which is further connected to a USB frame grabber card and the main computer The main computer is a 2 core powerful PC with a GPU that can be used for the intensive image processing computation The computer and the Sbot control board are connected using a serial port or
76. tislava Slovakia Sept 19 2010 Senzoricky syst m robota tvo optick d lkom r od firmy SICK LMS 221 barevn kamera SONY FCB IX11AP inerci ln jednotka s ipem ADIS163262 a GPS p ij ma s ipem SIRF3 D le je robot vybaven odometrem a elektro nick m kompasem CMPS03 Druh robotick platforma se jm nem Quido vych z z model sk ho pod vozku 1 10 s pohonem na v echna ty i kola Na voz tko byl vyroben hl nikov r m kter je s podvozkem pevn spjat a jen dovoluje uchycen i relativn t k ch senzor pomoc metrick ch roub Senzorick syst m vozidla Quido je obdobn jako u vozidla Karl k layer local position map matching ooos3 a s0os global position I Obr zek 1 Mechanick konstrukce robot Karl k a naviga n architektura robota 3 Naviga n architektura Autonomn zen mus b t spolehliv robustn v i chyb m a mus b et v re l n m ase Navr en naviga n architektura pokr v tyto po adavky pomoc distribuovan ch softwarov ch modul Architektura vych z z paraleln ho et zen ho zpracov n a proto je komunikace mezi moduly odli n vrstvy se azen a bufferovan viz obr 1 Vstupn data senzor jsou postupn transformov na do intern ch model pl n innosti a nakonec konkr tn ch pohybov ch kon Koncept naviga n architektury sest v ze t in cti modul b
77. to be more promising until it finds a heading where the module confidence 15 sufficiently high again If such heading is not found the robot attempts to return back in the reverse direction as it arrived to the problematic location possibly moving in the reverse of the planned direction on the map After returning back a short distance it retries The retries are repeated several times while gradually extending the back up distance If all attempts to pass the problematic location fail the planning module is asked to generate a different path The controller is arranged in a behavior based manner individual behaviors are developed and tested independently before they are integrated in a common controller V PATH RECOGNITION Our goal was to use artificial neural networks in order to help the robot navigate and stay on the path We obtained many images from a park with trails and we have manually marked the regions in these images that correspond to the traversable path This input was used to train the neural network a standard multi layer perceptron to recognize the path See figure 5 for an example of such manually classified image MEN 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 Figure 5 Manual preparation of training images Sending the whole image to the network as the input would obviously be infeasible Instead we first tried to scale the image to a lower resolution of 400x300 pixels and divide it
78. which are all part zs 5 Robotour Workshop Bratislava Slovakia Sept 19 2010 of a set from Parallax The aluminium framework allows mounting a rain shield for the computer and the camera when necessary Figure 1 3D Model of the robot showing main parts In real implementation we have mounted only one caster wheel as it proved to be sufficient and allowed more accurate control Figure 2 The resulting constructed robot from the side front and back The control electronics 15 installed under the PC The robot has already been tested in outdoor settings and has traveled a distance of several km III HARDWARE ARCHITECTURE The robot is propelled by two 12V DC motors with built in transmission rotating at up to 150 rpm and consuming 1 5A at no load The encoders with 36 ticks per rotation are used for speed and position feedback and are equipped with on board microcontrollers that are directly connected to the motor drivers HB25 supplying them with the proper PWM signal to keep the requested speed In this way the main microcontroller board which is the SBot control board designed in our group originally for SBot mobile robot is freed from the low level motor control and dedicates this task to both of the encoders that have an implementation of a standard P proportional controller and are connected using the same 1 wire serial bus Unfortunately we found that the original firmware for the encoders supplied by Paralla
79. x did not satisfy our needs for several reasons Most importantly the encoders were not designed for dynamic change of speed but only for simple positional commands that accelerate from zero speed to a fixed predefined speed and then decelerate after traveling the required distance They do not allow to change the speed in the middle of such positional command However movements where the speed and rotation is changed arbitrarily at any time are required in the Robotour task where the robot has to 16 dynamically respond to the visual feedback when it has to align its movement with the shape of the path Fortunately Parallax makes the source code for the encoders firmware available and thus we could modify it to suit our application and support immediate smooth changes of the instant speed The obstacles are detected using the standard SRF 08 and Maxbotix LV ultrasonic distance sensors that are connected to the main control board Outdoor robots are typically equipped with a global positioning device 1 e GPS and it is the case for our robot too Information from the GPS module that is connected directly to the main computer using USB port however is not so reliable due to atmospheric and other occlusions and serves only as a guidance for map localization It is confronted with visual input and complemented by the current heading obtained from compass sensor The compass sensor is part of the complex 9 DOF IMU sensor
80. zen posl ze pou iteln 1 na re ln m vozidle Toto modelov vozidlo je pracovn pojmenov no Ouido S vozidlem se t m astnil sout e Robotour 2010 Technick proveden voz tka Jako z klad pro vytvo en pohybliv ho vozidla modelu je pou it model sk podvozek 1 10 kter je p vodn ur en pro modely se spalovac m motorem Podvozek m pohon na v echny 4 kola dobr odpru en a byl upraven na elektrickou pohonnou jednotku Pro bezpe n uchycen sn ma na vozidle byl do vozidla p id n kovov hlin kov r m Na n m jdou p pravky s navrtan mi otvory umo uj c p ichycen 1 t k ch sn ma pomoc metrick ch roub Kovov r m je pevn spojen s autem zp sobem kov plast kov dic syst m Pro zen jsou vyu ity 1 2 notebooky s p slu n mi rozhran mi USB RS232 USB RS422 USB I2C Notebooky funguj pouze na svoje baterie Po ta e jsou spojeny v lok ln ethernet siti p es mal LAN a Wifi router V sou asn dob v konnostn posta uje pro kompletn zen vozidla jedin notebook druh se pou v v p pad verze d kladn j ho zpracov n videa Senzory Na vozidle jsou instalov ny n sleduj c senzory Optick radar Lidar LMS 100 p ipojen do notebooku p es TCP IP GPS pro ur en glob ln polohy e IMUl inerci ln jednotka obsahuj c 3D akcelerometry a 3D gyro p ipojen k notebooku p es CAN bus a RS232 V sou

Proceedings of the 5th Robotour Workshop

Contents

Download Pdf Manuals

Related Search

Related Contents