Final Report - Machine Intelligence Lab
Contents
1. The CMUcam comes with a Graphical User Interface written in Java Using this interface I tested the various capabilities of the camera I was able to track various colored object such as my hand and a blue ball The camera also dumped the mean color and the variance Dumping frames is also possible and I used this feature to properly focus the lens Once I had the camera in focus and was able to see data with the CMUcamGUI I began interfacing the camera with the microprocessor board To interface the CMUcam I found EEL 5666 Intelligent Machines Design Lab Page 24 of 38 Fall 2002 Final Report 12 13 2002 that removing the max232 chip and taking nor level shifted serial data from the camera to the non level shifted pins of the ATMega323 worked the best The following psuedocode is what I used to test the CMUcam Reset CMUcam Put camera into poll mode and turn on raw data mode for returned packets from camera Call Track Window LOOP START Call Track Color If of Pixels gt 250 And Confidence gt 50 Then Go Backward If Middle Mass X 55 And Confidence 25 Then Go Right If Middle Mass X 35 And Confidence 25 Then Go Left If of Pixels lt 120 And Confidence gt 25 Then Go Fwd GOTO LOOP START To test this algorithm I used the setup shown in the image below What I discovered when testing this code was that lighting is vital to the success of the CMUcam Standard incandescent lighbulbs worked ok but florescent2. Florescent lighting seems to work best Overall the project was a success and the objective was achieved EEL 5666 Intelligent Machines Design Lab Page 5 of 38 Fall 2002 Final Report 12 13 2002 Introduction I came up with my idea for Chester the autonomous waiter bot while I was enjoying one of my favorite pastimes being a couch potato I thought to myself wouldn t it be great 1f I had a robot that was capable of responding to me and serving me drinks and snacks without any effort on my own part This report will outline my proposal for an autonomous waiter bot that will do exactly what I dreamed of a robot capable of responding to a human request for service finding the human and providing a snack The brain of Chester is the Atmel AVR ATmega323 microcontroller It is the duty of the ATmega323 to control the actuation of the robot voice activation obstacle avoidance and the image processing features of the robot Voice activation was realized with the Voice Direct 364 Kit while image processing and human recognition was achieved with the CMUcam developed by researchers at Carnegie Mellon University EEL 5666 Intelligent Machines Design Lab Page 6 of 38 Fall 2002 Final Report 12 13 2002 Integrated Systems Left Wheel Servo Right Wheel Servo Atmel ATmega323 iu yer Mega A VR Development Board Port DO and D1 UART Figure 1 Integrated Systems of Chester Chester is comprised of fo
3. 9 gt 25 drive 1 0 Left if cpacket 8 lt 120 amp amp cpacket 9 gt 25 drive 1 1 Fwd if cpacket 9 lt 25 amp amp skip drive 0 0 Stop delay_ms 10 disable_pwm delay_ms 25 Timer 1 overflow interrupt service routine interrupt TIM1_OVF void timerl ovf isr void EEL 5666 Intelligent Machines Design Lab Page 32 of 38 Fall 2002 Final Report 12 13 2002 if camera init initiated return if count 0 check camera if homelock count 3 return check_ir bumper if sense 1 drive speed temp Lspeed temp r delay ms 175 random 7 42 sense 0 count 3 pragma savereg interrupt ADC_INT void adc_isr void asm push r26 push r27 push r30 push r31 in 130 sreg push r30 endasm register static unsigned char input_index 0 Read the AD conversion result adc data input index ADCH Select next ADC input if input index gt LAST ADC INPUT FIRST ADC INPUT input index 0 ADMUXZ FIRST ADC INPUTIADC VREF TYPE input index Start the AD conversion ADCSRI 0x40 asm pop r30 out sreg r30 pop r31 pop r30 EEL 5666 Intelligent Machines Design Lab Page 33 of 38 Fall 2002 Final Report 12 13 2002 pop r27 pop 126 endasm sese sk sk ok ok ok ok o o o o o k oe k oe oe oe oe oe o oe k k k k oe oe sese k kk k k k k k k K k k K Function drive Inputs Two integers between 2 and 2 representing the speed of the motors A
4. serving them a gumball The robot is also capable of chasing around a red ball on command An Atmel AVR ATMega323 is used to control the behaviors of Chester Chester sees his environment with four types of sensors including bump switches IR detectors a CMUcam and a Voice Direct 364 Kit These sensors are easily interfaced with the A TMega323 with plenty of room to spare The mobile platform for chester is made up of a 9 inch diameter circle supported by two 3 wheels and a rear castor A bridge extends across the center of the platform and several devices are conveniently mounted on this bridge including the voice kit speaker microphone and CMUcam Actuation for the robot is achieved by two hacked Futaba 83003 servos that provide 42 oz in of torque Timer 1 output compare is in PWM mode to generate the necessary signals to drive the robot The bump sensors and IR detectors proved ample for the robot to perform obstacle avoidance However objects not directly in front of the robot and objects extremely close present problems for the IR The bump switches compensate for this but the robot could use more of them The voice recognition kit is about 60 accurate if trained in the used environment and if the speaker is consistent in his her speaking of the key words The CMUcam works well if used properly and can detect bright colored objects from distances up to 9 feet away However lighting plays a large factor in the success of this sensor
5. you simply use the printf function For example to send put the camera into poll mode you would just write the C code printf PM I r It is important however to delay between commands to give the camera ample time to return from a command One thing to note about Codevision is that the evaluation version is free but if you want write larger programs you ll need the full version Since the printf command uses stdio h I ran into this problem very late into the semester Luckily I was able to share a license with my workplace I probably would recommend sticking with GCC and AVR Studio unless you are willing to pay 150 for a Codevision license Camera settings that were altered were polling mode and raw data mode The camera uses polling mode which sets the camera to send only one data packet back after each command rather than a constant stream of data Raw data mode is used to put the returned data packets into raw data rather than printable ASCII characters Raw data mode is nice because it allows the data to be easily processed The two main commands that will be used are the Track Window TW and Track Color TC commands TW used with no arguments is used to adjust the camera to the color found at the center of the tracking window The data obtained from the TW command is sent to subsequent calls to EEL 5666 Intelligent Machines Design Lab Page 17 of 38 Fall 2002 Final Report 12 13 2002 TC with no arguments The TC command return
6. 2002 Final Report 12 13 2002 Front Figure 2 Bump Switch placement on mobile platform To save analog ports I used a voltage divider network and sent its output to Port B2 or channel 2 of the a d converter Table 1 summarizes the different digital values for the bump sensors and what they mean To make obstacle avoidance more robust adding a left and right bump sensor may be necessary Voice Recognition EEL 5666 Intelligent Machines Design Lab Page 14 of 38 Fall 2002 Final Report 12 13 2002 For voice recognition I used the Voice Direct 364 VD364 kit from Sensory Inc The kit has two main modes slave mode and stand alone mode Slave mode allows the microcontroller to have complete control over the entire operation of the voice kit However slave mode require that the voice kit use the Serial Peripheral Interface SPI via a UART This is problematic b c the CMU cam will use the UART In addition the complexity of programming for the voice kit is exponentially greater than stand alone mode Because of its simplicity I ve decided to use stand alone mode for Chester In stand alone mode there are three different options e Speaker Dependent e Single Word Continuous Listening e Multi Word Continuous Listening Chester will use single word continuous listening In this mode the voice kit listens for a single gateway word which will activate listening for any of 15 words which you can train When a word is recognized i
7. 66 Intelligent Machines Design Lab Page 37 of 38 Fall 2002 Final Report 12 13 2002 j GLOBAL VARIABLES void init camera void printf RS r delay_ms 100 printf r delay_ms 100 printf PM I r delay_ms 100 printf RM 3 r delay_ms 5000 while adc_data 2 88 while UCSRA 7 temp UDR flush the receive buffer printf TW r delay_ms 2000 while adc_data 2 88 wait for back bumper press to begin camera_init 1 PORTC PORTC while UCSRA 7 temp UDR flush the receive buffer void main void LOCAL Variables chester init Initialize A D timers interrupts and USART disable pwm delay ms 1500 Delay to give the A D and other systems a chance to warm up asm sei init_camera while adc data 3 OxEO Wait until voice command given initiated 1 drive 1 1 begin moving delay_ms 1000 while 1 EEL 5666 Intelligent Machines Design Lab Page 38 of 38 Fall 2002 Final Report 12 13 2002 random num random 50 while random num 0 amp amp homelock drive 2 2 delay ms 100 if homelock drive random 4 2 random 4 2 change directions drive 1 1 spin in place looking for objec random num random 15 while random_num 0 amp amp homelock delay_ms 100 if stopped 1 while adc_data 3 lt OxEO amp amp adc data 4 OxEO stopped 0 if adc_data 3 gt O0xEO homelock 0 drive 1 1 delay_m
8. David Winkler 12 10 02 EEL 5666 Intelligent Machines Design Lab Dr A Antonio Arroyo TAs Uriel Rodriguez Jason Plew Final Report EEL 5666 Intelligent Machines Design Lab Page 2 of 38 Fall 2002 Final Report 12 13 2002 Table of Contents LE JADSU 6Lo o d s s ea 3 II Executive Summary eese 4 III Introduction nnn 5 IV Integrated System eeeeesses 6 V Mobile Platform esses 7 VL ACUOLO e are o e E Ea s 9 VI Se OFS aiei a a aa etai 9 VIII Behaviors nnn III 25 IX Conclusion een 28 EEL 5666 Intelligent Machines Design Lab Page 3 of 38 Fall 2002 Final Report 12 13 2002 Abstract This document is the final report for Chester the autonomous waiter bot In this report I intend to discuss my goals for Chester and what I achieved including the robots behaviors and the sensors and electronics that were used The ultimate goal for Chester was to use voice recognition to respond to a human request for service find the human target and serve him a drink and a snack The project met all these objectives EEL 5666 Intelligent Machines Design Lab Page 4 of 38 Fall 2002 Final Report 12 13 2002 Executive Summary This report describes the design and implementation of an autonomous mobile waiter robot named Chester The robot is capable of responding to human voice command finding the human and then
9. EL 5666 Intelligent Machines Design Lab Page 29 of 38 Fall 2002 Final Report 12 13 2002 pursue other projects and I now feel well equipped to solve more difficult engineering problems Appendices The following is the final version of the code for Chester The program was written using CodevisionAVR C compiler 78 7 7 2 E sk ok ok ok ok o o o k oe ole oe oe oe oe oe oe k k kk k kk k k kkk k kk k k K k k K K Project Chester Final Code Version 3 0 Date 12 05 2002 Author David Winkler Chip type ATmega323 Clock frequency 6 000000 MHz Internal SRAM size 2048 External SRAM size 0 Data Stack size 512 PEOR ERE GE GE SEE REESE sese se oe oe oe ole ok oe oe olere k k k k he k oe o kk kk k k k k k k kkk Esse 2g sk ok ok ok ok o o o o ole o ole oe oe oe oe oe k o o kk k k kk k kk k k kk k K k Trying to get the robot to spin in place to look for red ball also attempting to get the robot to stop once it thinks it finds the ball spin in place and then back up towards ball to dispense gumballs PEOR RE GERE GE RE REESE SE sese se sese k oe oe o oe oe oe k k k k he oe oe o seo k k k k k K k k K K INCLUDES include mega323 h include lt delay h gt include lt stdio h gt CONSTANTS define ADC_VREF_TYPE OxEO define L_EYE TOLERANCE 50 define R EYE TOLERANCE 50 define FIRST ADC INPUT 0 define LAST ADC INPUT 4 define BACK BUMP 88 define FLEFT BUMP 131 EEL 5666 Intelligent Machines Desi
10. MUNI d ui m n First I will discuss the tests I did on the IR sensors To test the IR sensors I wrote a program which would get data from the IR sensors and print them to the computer screen EEL 5666 Intelligent Machines Design Lab Page 19 of 38 Fall 2002 Final Report 12 13 2002 through the UART I placed several different objects in front of Chester and read off the values of the IR sensors for different distances I obtained the following results IR Readings for Shoe Object D E a o tc E E E 6 7 5 95 11 13 14 16 18 5 20 23 Right IR Sensor Distance Inches a Left IR Sensor EEL 5666 Intelligent Machines Design Lab Page 20 of 38 Fall 2002 Final Report 12 13 2002 IR Readings for Box Object 100 80 aN Right IR Sensor s Left IR Sensor 60 Atmel A D Values 40 20 0 N V e oo 5 do Xx A v oq yo Distance Inches IR Readings for Dresser Object e Right IR Sensor s Left IR Sensor Atmel A D Value 9 5 11 5 14 16 195 21 5 24 Distance Inches EEL 5666 Intelligent Machines Design Lab Page 21 of 38 Fall 2002 Final Report 12 13 2002 The program used to test the IR sensors was also used to test the bump sensors The code for this program is provided in the appendices The following voltage divider circuit was used to distinguish between the different bump sensors The following table describes the analog values obtained
11. Mobile Platform For the mobile platform I ended up modifying the Talrik Pro platform because of my limited experience and to keep things as simple as possible The platform is circular in shape and is 9 in diameter It rolls along on two Du Bro 3 diameter wheels powered by two hacked Futaba 3003 servos Spanning across the center of the platform is a bridge on which is mounted the voice kit and the CMU cam A small bump skirt surrounds the EEL 5666 Intelligent Machines Design Lab Page 8 of 38 Fall 2002 Final Report 12 13 2002 platform and provides a contact mechanism with four bump switches placed strategically around the periphery of platform The gumball dispensing machine is placed on the rear of the platform and is secured with velcro A small caster wheel is mounted on the rear of the platform For power I used 8 1700mAh NiMH batteries The batteries were secured to the bottom of the platform with a velcro tie Figures 2 3 and 4 show the layout of the platform Figure 2 Rear view of Chester EEL 5666 Intelligent Machines Design Lab Page 9 of 38 Fall 2002 Final Report 12 13 2002 Front Bump 2 Figure 4 Front view EEL 5666 Intelligent Machines Design Lab Page 10 of 38 Fall 2002 Final Report 12 13 2002 Actuation For movement the robot rolls around on two 3 Du Bro wheels that are powered by two hacked Futaba S3003 servos connected to the ATmega323 s PWM channels Timer Counter 1 was used to generate the PWM
12. Only two noticeable speeds were generated by varying the PWM castor wheel on the back provided support for the platform but it kept the robot from moving in a straight line Perhaps a ping pong ball or something else would support the platform and not interfere with the direction trying to be imposed by the movement of the wheels I also had a problem with the hacked servos Initially they were calibrated to not move for a given PWM but they eventually became misaligned and would move slightly when given the stop PWM The servos unfortunately were glued to the platform and could not be removed To compensate for this I just disabled the PWM whenever the robot came to a stop The CMUcamalso has a servo port capable of using a servo to track a color This could add another level of actuation to the robot However this would add complexity to the object tracking code so this feature was not used Sensors There are four different types of sensors used in Chester 1 Infrared IR detectors EEL 5666 Intelligent Machines Design Lab Page 11 of 38 Fall 2002 Final Report 12 13 2002 2 Voice Recognition 3 Bump Switches 4 Image Processing Camera CMUcam IR Sensors Chester uses two Sharp GP2D12 IR detectors mounted on the top of the platform for obstacle avoidance The GP2D12 sensors measure distances of 4 to 30 and return an analog voltage The analog voltage from Chester s IR sensors are sent to PortBO and PortB1 which are chan
13. for the different combinations of bump switches Back 8B Front and Front Left 144 Front and Front Right 62 0 0 O 88 Front Right 144 1 0 62 EEL 5666 Intelligent Machines Design Lab Page 22 of 38 Fall 2002 Final Report 12 13 2002 For some of the bump switch combinations the A D reading would not stabilize at any one value Instead it would toggle between the value 1 This is probably because of the lack of resolution for the A D result The A D was configured for only 8 bits of resolution and the bump sensors were sampled at 23 kHz The circuit used to test the voice kit is shown below Single Word CL Configuration Tcp wow of tho Voca Dict 354 EEL 5666 Intelligent Machines Design Lab Page 23 of 38 Fall 2002 Final Report 12 13 2002 This circuit configures the voice chip for single word continuous listening mode In this mode the circuit is continuously listening for the gateway word Once the gateway word is spoken the voice kit listens for any of 15 different words To test this configuration I trained the word Chester for the continuous listening and trained the speaker dependent words come here and go away These I trained at three different distances and then I tested the recognition accuracy at various positions The results of these tests are given by the following table Accuracy at 3 ft Accuracy at 6 ft Accuracy at 9ft Accuracy at 12 ft Note Accuracy based on 10 trials
14. gn Lab Fall 2002 Final Report define FRIGHT BUMP 24 define CENTER BUMP 46 GLOBAL VARIABLES unsigned int adc dat LAST ADC INPUT FIRST ADC INPUT 1 unsigned int count 5 short int speed temp 1 0 short int speed temp r 0 short int speed val 1 20 short int speed val r 0 int temp bit sense 0 bit initiated 0 unsigned char random num unsigned char cpacket 10 bit homelock 0 int count120 bit camera_init 0 bit ball_chasing 0 bit stopped 0 bit skip 0 void check ir void void drive int int int random char void bumper void void disable pwm TCCRIA 0x01 void enable_pwm j TCCRIA 0xA1 void check_camera void PORTC 0 0 count1 0 printf TC r Page 30 of 38 12 13 2002 EEL 5666 Intelligent Machines Design Lab Page 31 of 38 Fall 2002 Final Report 12 13 2002 while count1 11 cpacket countl 2getchar count PORTC 0 1 if cpacket 9 gt 40 homelock 1 if homelock 45 is H center if cpacket 8 gt 250 amp amp cpacket 9 gt 50 drive 1 1 if ball_chasing delay_ms 300 drive 1 1 delay_ms 1600 drive 1 1 delay_ms 400 drive 0 0 delay_ms 10 disable_pwm stopped 1 robot stops and waits for voice command else delay ms 300 drive 0 0 delay ms 10 disable pwm if cpacket 2 gt 55 amp amp cpacket 9 gt 25 drive 0 1 Right if cpacket 2 lt 35 amp amp cpacket
15. lighting such as that in the lab works much better In the lab environment Chester recognizes objects EEL 5666 Intelligent Machines Design Lab Page 25 of 38 Fall 2002 Final Report 12 13 2002 from 7 9 feet away while in my apartment Chester will recognize objects from only 3 5 feet away These results were much better than several other people in the class perhaps b c I bought my camera from Acroname rather than Seattle Robotics The camera from Acroname comes with an IR filter already installed while I believe it s a special order from Seattle Robotics Behaviors Chester will perform the following behaviors e Listening behavior e Obstacle avoidance behavior human tracking behavior e Human tracking behavior e Dispensing pause nehavior e Ball chasing behavior The following flow chart relates all the behaviors and gives an overview of the operation of Chester EEL 5666 Intelligent Machines Design Lab Page 26 of 38 Fall 2002 Final Report 12 13 2002 Listening For Serve Initial State Command Human Finding Obstacle Avoidance Serve human and Leave wait for voice Ball Chase Behavior Command command Listening behavior When Chester is first turned on it will immediately enter listening behavior In this state Chester is awaiting a voice command to commence serving Once the human gives the voice command to serve Chester will enter Obstacle Avoidance Human tracking behavio
16. mers LLLLLLLLLLLLLL 2 2 k k 2 2 2 2h k k k k k k k k k k k kk k k kk k k K k k K K void chester_init void Input Output Ports initialization Port A initialization PORTA 0x00 DDRA 0x00 Port B initialization PORTB 0x00 DDRB 0x00 Port C initialization PORTC 0x00 DDRC OxFF Port D initialization PORTD 0x00 DDRD 0x30 EEL 5666 Intelligent Machines Design Lab Page 36 of 38 Fall 2002 Final Report 12 13 2002 Timer Counter 0 initialization Clock source System Clock TCCRO 0x01 TCNTO 0x00 Timer Counter 1 initialization Clock source System Clock Clock value 23 438 kHz Mode Ph correct PWM top 00FFh OC1A and OCIB output Non Inverted Input Capture on Falling Edge TCCRIA 0xAI TCCR1B 0x04 TCNT1H 0x00 TCNT1L 0x00 OCR1AH 0x00 OCR1AL 0x18 OCR1BH 0x00 OCR1BL 0x18 Timer 1 Overflow Interrupt Enabled TIMSK 0x04 USART initialization Communication Parameters 8 Data 1 Stop No Parity USART Receiver On USART Transmitter On USART Mode Asynchronous USART Baud rate 9600 UCSRA 0x00 UCSRB 0x18 UCSRC 0x86 UBRRL 0x26 UBRRH 0x00 Analog Comparator initialization Unused ACSR 0x80 SFIOR 0x00 ADC initialization ADC Clock frequency 93 750 kHz ADC Voltage Reference AREF pin Internal 2 56 V Only the 8 most significant bits of the AD conversion result are used ADMUX ADC_VREF_TYPE ADCSR 0xCE EEL 56
17. nels 0 and 1 of the Atmel ATMega323 A D converter The analog voltage is then converted into a number between 0 and 255 and various decisions can be made based on the converted values The IR detectors are placed at 15 degree angle inward and are placed on opposite sides of the platform For two IR sensors this arrangement works well as Chester has been able to avoid obstacles with the current obstacle avoidance routine that I have written The only problem I ve had with this configuration is in the case when objects are not directly in front of Chester yet they are still in the path of the robot In the current configuration of the IR sensors it will not be possible for this situation to be detected However with the addition of a couple of bump sensors this problemcan be eliminated The following diagram shows the placement of the IR detectors EEL 5666 Intelligent Machines Design Lab Page 12 of 38 Fall 2002 Final Report 12 13 2002 SU Pe nd aye aud Figure 1 IR Placement There may be a more optimal placement of the IR sensors I ve been experimenting with different configurations to see if it improves my obstacle avoidance algorithm Bump Sensors Four bump switches are placed on the perimeter of Chester s platform These provide another means of obstacle avoidance given that the IR sensors fail The following is the placement of the bump sensors on the platform EEL 5666 Intelligent Machines Design Lab Page 13 of 38 Fall
18. o Chester he will turn 360 degrees looking for the object If Chester still does not see the object he will move in a random pattern around the room and then EEL 5666 Intelligent Machines Design Lab Page 28 of 38 Fall 2002 Final Report 12 13 2002 spin in place again looking for the object This will continue until Chester finds the desired object Once Chester locates the object he will move towards the object until he reaches a certain distance Once Chester has reached the desired distance from the brightly colored object he will enter the dispensing pause behavior In this behavior Chester rotates 180 degrees and aligns the gumball machine with the tracked object Dispensing pause behavior In this behavior the robot will be positioned in front of the brightly colored object The human can now dispense the gumballs from the robot at his her leisure Once the human has been served he she gives Chester either the leave command or the ball chase command If the ball chase command is given Chester turns around faces the ball and will chase the ball around Ball Chasing Behavior In this behavior Chester will turn around and face the ball or object and will follow the object as long as the object is held in view of the camera tracking window Conclusion This concludes my final report for Chester the autonomous waiter robot The robot worked as specified and was a success This project has given me the confidence to E
19. r While in this behavior Chester is calibrated to his environment To calibrate the camera the object to be tracked is placed about 5 inches away from the CMUcam EEL 5666 Intelligent Machines Design Lab Page 27 of 38 Fall 2002 Final Report 12 13 2002 The rear bumper is then pressed and the camera captures the color of the object with the track window command It is also very important in this behavior to train the voice sensor to the environment that it will be used in If the voice sensor is not trained the robot likely will not operate properly Obstacle Avoidance Human tracking behavior The obstacle avoidance behavior was taken from Eric Donnelly s WOMAN Robot final report Spring 2002 The algorithm based on random numbers seemingly gives the robot a mind of its own when avoiding obstacles Chester will turn in random directions at random speeds for random amounts of time This makes the robot more interesting and gives the robot better coverage of an area While in obstacle avoidance mode Chester s senses are checked in a priority fashion The following is the order of priority from highest to least e Object Tracking e Bump Switches e IR Detectors While Chester is in obstacle avoidance mode he is collecting data from the IR detectors and bump sensors to avoid obstacles Meanwhile Chester uses the CMUcam to try to track the brightly colored object placed at the feet of the human If the object is not immediately visible t
20. s 1000 else if adc_data 4 gt OxE0 skip 1 ball_chasing 1 drive 2 2 delay_ms 400 drive 1 1 delay_ms 400 drive 0 0 skip 0
21. s a data packet which is then used to make decisions as to how to position the robot The following is the structure of a packet returned from TC Type M packet M mx my x1 yI x2 y2 pixels confidence r The M indicates that the camera is returning a middle mass TC packet Mx is the middle mass x coordinate of the tracked region and My is the middle mass y coordinate The next four bytes give the coordinates of the tracked rectangle Pixels is the of pixels in the tracked region capped at 255 Confidence indicates how well the camera has a lock on the desired color A confidence 50 indicates a good lock while a confidence 10 indicates a poor lock Chester uses an interrupt service routine to periodically check the confidence level for color tracking data and is able to determine if the colored object is nearby For proximity measurements of pixes will be used to determine if the object is close or far away to Chester Mx will be used to align the robot with the object and will keep the robot from veering of to one side For more information see the CMUcam user manual http www 2 cs cmu edu cmucam Downloads CMUcamManual pdf It describes the entire serial command set and what all the extrapolated data means EEL 5666 Intelligent Machines Design Lab Page 18 of 38 Fall 2002 Final Report 12 13 2002 Experimental Layout and Results The following PCB is what I used to connect the sensors to my microprocessor board EIE
22. speed gt 0 is forward while a speed 0 is backward The greater the magnitude of the inputs the faster the motors will turn Outputs none Notes A 100 ms delay is implemented when changing motor speeds to protect the motors from burnout LDLLLLLLLLLLL 2 2 2 E 2 2 2 2 2 2 2 k 2 k k k he fe k fe k LLL k kk k k k k k k k kk void drive short int speedl short int speedr enable pwm if speedl 2 can t overdrive servos speed 2 8 sec max total to change speeds if speedl 2 speed 2 if speedr gt 2 speedr 2 if speedr lt 2 speedr 2 while speed val 1 speed ll speed_val_r speedr if speed val 1 gt speedl OCRIBL OCRIBL 1 speed val L if speed val 1 lt speedl OCRIBL OCRIBL 1 speed_val_l if speed_val_r gt speedr OCRIAL OCRIAL 1 speed val r if speed_val_r lt speedr OCRIAL OCRIAL 1 speed_val_r delay_ms 25 delay on each increment to protect motors EEL 5666 Intelligent Machines Design Lab Page 34 of 38 Fall 2002 Final Report 12 13 2002 j int random unsigned char range generates random numbers between 0 254 range return TCNTO range void check_ir void if adc_data 1 gt LLEYE_TOLERANCE ll adc data 0 gt R_EYE_TOLERANCE sense 1 temp random 2 2 random speed between 2 and 1 if adc_data 1 gt adc data 0 10 speed temp l temp speed temp r temp random 3 1 always point awa
23. t activates an output pin for one second This output will be a digital I O on the Atmel chip and will be sampled by software to detect the presence of a command EEL 5666 Intelligent Machines Design Lab Page 15 of 38 Fall 2002 Final Report 12 13 2002 The circuitry for the voice circuit is very simple as can be found in the voice direct kit manual Using Protel I have prototyped this circuit and mated it with the VD364 kit The following figure shows the final product Figure 6 Voice Circuit One thing to note about the VD364 is the choice of word for the recognition set The performance of the kit depends highly upon what is chosen for a command set The manual recommends choosing words that sound dissimilar and that vary in syllables Also intonation of the voice when training the kit is very important I ve had situation where I trained a the kit in a monotone voice but when I go to give it a command I get excited and change the intonation of my voice The kit then has a hard time recognizing the word as part of its command set EEL 5666 Intelligent Machines Design Lab Page 16 of 38 Fall 2002 Final Report 12 13 2002 CMU Cam Chester uses the CMUcam to find the human giving it the command Using the CMU Cam s color tracking abilities Chester will find the human which is giving it the command The CMUcam communicates with the Atmel microprocessor via the USART Using Codevision C compiler this is made easy To send a command
24. ur main systems e Object avoidance system e Voice recognition system e Human recognition system e Object dispensing sytem Three of those systems are controlled by the onboard microcontroller The microcontroller I used for control was the Atmel AVR ATmega323 EEL 5666 Intelligent Machines Design Lab Page 7 of 38 Fall 2002 Final Report 12 13 2002 Following the KISS principle the object dispensing system was controlled by the microcontroller The human is able to dispense the objects he she desires and notify the robot via a voice command that they have taken what is desired from the robot The Object avoidance system is comprised of two Sharp GP2D12 IR detectors and bump switches The Voice recognition system is built upon the Voice Direct 364 www voiceactivation com Some of the features of the Voice Direct 364 courtesy of the www sensoryinc com are e Seaker Dependent and Continuous Listening speech recognition technologies e Minimal external components e Recognizes up to 60 words or phrases in slave mode or 15 in stand alone mode broken in to 1 2 or 3 sets e Over 99 recognition accuracy with proper design e Phrase recognition up to 2 5 seconds e User friendly speech prompting The human recognizing system was achieved by using the CMUcam The CMUcam is capable of detecting and tracking bright colors The camera detects a human by tracking a brightly colored object placed next to the human requesting service from Chester
25. y from obstacle else if adc_data 0 gt adc_data 1 10 speed temp l temp random 3 1 speed_temp_r temp always point away from obstacle else speed temp 1l random 4 2 straight ahead turn in place speed temp r speed temp I void bumper void if adc_data 2 0 Il adc_data 2 1 return sense 1 while 1 if adc_data 2 45 ll adc_data 2 46 Il adc_data 2 47 II adc_data 2 143 ll adc_data 2 144 ll adc_data 2 145 II adc_data 2 61 ll adc_data 2 62 Il adc_data 2 63 front drive 1 1 back up delay_ms 300 speed temp 1l random 4 2 turn in place if straight speed temp r speed temp I forward break else if adc_data 2 130 Il adc_data 2 131 ll adc_data 2 132 left front speed temp I 1 right will reverse harder EEL 5666 Intelligent Machines Design Lab Page 35 of 38 Fall 2002 Final Report 12 13 2002 speed temp r 2 break else if adc_data 2 23 ll adc_data 2 24 Il adc_data 2 25 right front speed_temp_r 1 left will reverse harder speed temp l 2 break else if adc_data 2 87 ll adc_data 2 88 ll adc_data 2 89 forward speed_temp_r random 1 1 speed_temp_l random 1 1 break else return sese sese ok ok ok ok ok o o o o k o fe oe oe oe oe oe k o o k kk k k k obe obe k obe obe kk K k k K K Function chester_init Input None Output None Purpose Set up A D I O Ports UART Interrupts and Ti
Download Pdf Manuals
Related Search