A taxonomy and comparison of haptic actions for disassembly tasks
Contents
1. The haptics taxonomy provided a framework for exam ining maintenance actions Our results implied an execu tion time advantage for users executing combined force and torque tasks using the Phantom over non haptics Cyber Glove and SpaceMouse devices Generalizing this result it is possible that complex tasks could benefit from human feasibility analysis using haptics simulation and thus that selected task simulations can assist in developing valid service manuals The primary moderat ing factor in our experiments was the relatively low levels of force and torque reacted through the Phantom As users were more successful at simulated tasks using haptic feed back it is possible that more strenuous challenges would benefit from similar feedback but this would need to be tested in a more industrial strength haptics feedback en vironment The difficulty here is ensuring the safety of the subjects Recall that we categorized virtual task performance four ways 1 Physical performance of task in real environment 2 Interactive computer user with visual feedback only 3 Interactive computer user with visual and haptic feed back 4 Non interactive computational task analysis Significant benefits would accrue to system and task de signers if the fourth option were realizable Obtaining ge ometric models of aircraft systems is not too difficult and digital human models are commercially available But our experiments seem to show2. be unlikely that benefits would accrue for simpler cases Conversely were we able to detect benefits for haptics we would expect sim pler cases to inherit similar benefits 4 Experimental Procedure The virtual reality simulation that forms the basis for this experiment consisted of a series of actions the participants IEEE P COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE dl Force Only I Force Only II Torque Only I Torque Only II secede areas Fine motor control Significant arm strength Tactile finger pressure friction Cooperative two handed tasks Braced two handed tasks Manipulating a deformable object Tool assisted tasks Multi finger tasks Pushing a pin Pulling open a stuck access panel Pushing a button Lifting a bulky object Pushing two connectors together Pushing to create a Shape fuel bladder removal Interface to increase force per unit area hammer chisel Pushing multiple buttons at once Wiping off grease Sanding Loading grease into a hole Filing with a large file Holding a support while doing a significant arm strength task Holding a cloth steady as it flaps in the wind Interface to over come friction or to increase force per unit area plane crowbar Pulling a pin by its head Turning a dial Turning a stuck Inserting a small bolt Extracting a large threaded ro
3. radius 5 Statistics and Experimental Measures The main data point used for statistical analysis is the sum of the times taken for the individual actions This mea surement is the main dependent variable because it takes into account mistakes made and corrective steps taken i e you can t move to the next step without accurately com pleting the previous one so mistakes cause an increase in time The completion time does not include the time be tween steps 1 e finding and grabbing the tool that is vari able across the devices For each simulation the times for the individual actions were summed These sums were then averaged across the last four simulations The first two simulations were ex cluded from analysis because they served as training runs with experimenter input A one way analysis of variance ANOVA was performed to determine whether statistically significant differences existed among the three devices in terms of this average time Ratings from the questionnaire were entered into a sep arate ANOVA to examine differences in perceived ease of use of the devices The question was rated on a scale of 1 to T To ensure equivalence of subject variables across groups demographic data were analyzed via ANOVA or chi square tests These data included age gender and number of hours spent using a computer mouse in an average week All analyses were run with a level of significance set at p lt 05 6 Experime
4. that task performance depends on haptic feedback in a way that no existing computational model can accommodate Embedding human performance capability models into kinematic human models seems to be a necessary and desirable step toward automating the analy sis of maintenance tasks for service manual generation and validation 9 Acknowledgments This research is partially supported by Air Force F33615 99 D 6001 Delivery Order 8 and NASA 00 HEDS 01 052 References 1 N Badler R Bindiganavale C Erignac and P Vincent Technology assessment for automated technical order val idation Technical report Air Force Research Laboratory Deployment and Sustainment Division AFRL HES 2000 N I Badler C A Erignac and Y Liu Virtual humans for validating maintenance procedures Communications of the ACM 45 7 56 63 July 2002 3 G C Burdea Force and Touch Feedback for Virtual Reality John Wiley and Sons New York 1996 S Card W K English and B J Burr Evaluations of mouse rate controlled isometric joystick step keys and text keys for text selection on a CRT Ergonomics 21 8 601 613 Aug 1978 5 M R Cutkosky On grasp choice grasp models and the design of hands for manufacturing tasks IEEE Journal of Robotics and Automation 5 3 269 279 June 1989 6 M R Cutkosky and P K Wright Modeling manufactur ing grips and correlations with the design of robotic hands In Proceedings of the IE
5. through grip space such as using a lever device 1 e a wrench for leverage or turning a steering wheel e Force and Torque is a broad category containing any action that requires both force and torque to complete such as using a bayonet connector A couple of potential categories were omitted or com bined Requiring no force or torque is not particularly inter esting and was omitted The last category where both force and torque are required could have been further subdivided into four areas each of the two force types with each of the two torque types but we opted to combine them into one The haptic action task taxonomy is shown in table 1 The taxonomy helped determine which actions could be profitably simulated and compared in a virtual disassem bly environment We focused on the combination force and torque tasks to provide both utility and challenge to the experiment subjects Everyone has had experience with both screwdrivers and bayonet connectors such as a child proof medicine bottle cap yet the user s actions are mostly dictated by the physics forces and torques and geometry size shape turning angle of the object Our hypothesis is that the execution of those actions in a virtual environment would benefit from haptic feedback that is what we sought to test Since combinations of force and torque ought to be more complex than either alone were we to detect no sig nificant benefits to haptic feedback it would
6. A Taxonomy and Comparison of Haptic Actions for Disassembly Tasks Aaron Bloomfield Yu Deng Jeff Wampler Pascale Rondot Dina Harth Mary McManus Norman Badler University of Pennsylvania Center for Human Modeling and Simulation Air Force Research Laboratory Deployment and Sustainment Division AFRL HES 3GE Global Research aaronb cis upenn edu Abstract The usefulness of modern day haptics equipment for vir tual simulations of actual maintenance actions is exam ined In an effort to categorize which areas haptic simu lations may be useful we have developed a taxonomy for haptic actions This classification has two major dimen sions the general type of action performed and the type of force or torque required Building upon this taxonomy we selected three representative tasks from the taxonomy to evaluate in a virtual reality simulation We conducted a series of human subject experiments to compare user per formance and preference on a disassembly task with and without haptic feedback using CyberGlove Phantom and SpaceMouse interfaces Analysis of the simulation runs shows Phantom users learned to accomplish the simulated actions significantly more quickly than did users of the Cy berGlove or the SpaceMouse Moreover a lack of differ ences in the post experiment questionnaire suggests that haptics research should include a measure of actual per formance speed or accuracy rather than relying solely on subjective
7. EE International Conference on Robotics and Automation pages 1533 1539 1986 7 V Hayward and O R Astley Performance measures for haptic interfaces In G Giralt and G Hirzinger editors Robotics Research The 7th International Symposium pages 195 207 1996 8 S Jayaram U Jayaram Y Wang H Tirumali K Lyons and P Hart VADE A virtual assembly design environment IEEE Computer Graphics and Applications 19 6 44 50 Nov Dec 1999 J Payette V Hawyard C Ramstein and D Bergeron Eval uation of a force feedback haptic computer pointing de vice in zero gravity In Proceedings of the ASME Dynamcis Systems and Control Division volume 58 pages 547 553 1996 10 G Sclesinger Der mechanische aufbau der kunstlichen glieder the mechanical building of artifical limbs In M Borchardt editor Ersatzglieder und Arbeitshilfen fur Kriegsbeschadigte und Unfallverletzte Prosthetic and Oc cupational Help for War Victims and Accident Victims pages 321 699 Springer 1919 11 M A Srinivasan and C Basdogan Haptics in virtual envi ronments taxonomy research status and challenges Com puters and Graphics 21 4 393 404 July Aug 1997 12 C L Taylor and R J Schwarz The anatomy and mechanics of the human hand In Artifical Limbs 2 pages 22 35 1955 13 United States Air Force Technical Order 1F 16CG 2 28 G 20 1 Figure 2 14 1 Removal of Internal Fuel Tank Vent and Pressurization Va
8. ated actions significantly more quickly than did users of the CyberGlove or the SpaceMouse This could translate into lower training costs and or more effi cient training sequences The 3D graphics simulations and the user s ability to directly manipulate objects were com parable so we attribute the decreased performance times to two main factors The Phantom differs from the other two devices in sev eral ways including its use of haptic feedback It is possible that the haptic feedback contributes to faster performance although a more direct comparison of haptic feedback vs no haptic feedback both using the Phantom would offer more rigorous assessment of this variable IEEE P COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE The other factor is the increased precision that is inher ent in the Phantom The range of motion for the Phantom 1s limited to a specific volume the range of the Phantom arm which inherently maps onto the virtual world By compari son the SpaceMouse and CyberGlove users had more diffi culty moving to a specific location in the simulation The lack of differences in the post experiment question naire suggests that future research should include a measure of actual performance speed and or accuracy whether in simulations or in real life tasks rather than relying solely on subjective reports of a device s ease of use 8 Future Work
9. capability with a haptics enabled virtual validation environment The SMG program is rev olutionizing the way technical manuals are developed and providing a unique opportunity to validate the manuals be fore the system is built Our research is directly supporting the SMG effort by evaluating various input devices across representative maintenance tasks The overall goal of this study is to investigate and com pare virtual task validation with and without haptic feed back There are four general methods to test and validate task performance 1 Physical performance of task in real environment 2 Interactive computer user with visual feedback only 3 Interactive computer user with visual and haptic feed back 4 Non interactive computational task analysis Case 1 involves human action in a real system or full scale mock up Cases 2 and 3 are investigated here using three different devices Case 4 uses computation alone such as a robotics reachability analysis and digital human mod IE O COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE els and is not presently feasible primarily due to human variability for complex situations In any of these cases there are four possible general out comes A The task is physically possible and humanly possible by any typical maintainer B The task is physically possible but unreasonable to expect from a typical ma
10. connected to the fuel tank The coupling on the upper right side of the tank was removed next also with a wrench The elbow attached to the coupling rotated automatically once this coupling was removed The next two actions were the removal of bayonet connectors the first on the lower right of the tank and the second on the upper left of the tank The last actions were the removal of the three screws that held the tank on the base Figure 3 shows a completed simulation for the Phantom The base of the tank slides downward to indicate that the Figure 3 Completed simulation for the Phan tom simulation has been completed The removal of the last three screws is not in the original technical order but was included to provide the feeling of completing the task of removing the fuel tank These last three screws were not in cluded in the data analysis the first two screws were how ever Each participant ran the simulation on one of the three platforms The subjects were arbitrarily assigned to a de vice After signing the consent form the subject viewed two videos one that described the purpose of this research and what they were about to do and the second which de scribed how to use the particular device they were assigned The subjects then proceeded to run through the simulation a total of six times The first time was with an experimenter helping to guide them through the process Experimenter input was also permitted in the seco
11. d or bolt Wringing a towel Interface to increase torque hex screwdriver Turning a dial Using an X wrench Using a wrench turning a crank Dialing a rotary type phone Dumping a wheelbarrow load Stirring a viscous liquid Interface to increase torque via leverage wrench Turning a large Inserting a bayonet connector Pushing a heavy door while turn ing a lever latch Inserting a bayonet connector Pulling and twisting a piston from a cylinder Twisting wires while pulling the cable taut Interface to increase torque and force per unit area screwdriver Inserting a wing nut bayonet connector Table 1 Maintenance Task Taxonomy had to perform to complete a disassembly task The actions were using a screwdriver tool assisted torque I using a wrench tool assisted torque II and removing a bayo net connector fine motor control force amp torque although there are other categories it could fit in The virtual real ity simulation chained multiple instances of these actions together to create the disassembly task The scenario we used for our experiments was part of the removal of a fuel tank from an F 16 aircraft As the pur pose of our research was to evaluate the three devices and to use a range of actions from the taxonomy accuracy to the real task was not necessary Thus the simulation was based loosely on the actual Air Force Technical Order mainte nance
12. finger pressure friction These actions include anything that requires significant tactile feedback A pri mary purpose of haptics is to provide touch and force feed back but in these actions touch dominates One can push open a door without haptic feedback although the haptic feedback makes it feel more realistic Examples include pushing a button loading grease into a hole or dialing a rotary type telephone Cooperative two handed tasks These tasks include those actions that require the two hands to cooperate to achieve the desired effect Examples include pushing two connectors together using a two handed tool or lifting a bulky object Note that the actions that each hand performs might fit into another category as well and may fit into different categories These actions require multiple haptic devices along with the necessary computer support equip ment to operate them This often makes simulating these tasks difficult Braced two handed tasks These tasks require one hand to brace itself against an object while the other per forms the action Examples include holding a support while using a tool or bracing one hand against a wall to pull open a door The task performed with the one hand is often a task requiring significant arm strength as if it were not it would not require bracing These tasks are much easier to simu late than the cooperative two handed tasks as one does not need a separate haptic device on the bracin
13. g hand Only a rigid prop for the bracing hand is needed though placement of the prop will need to be customized to the geometric con figuration of the task Manipulating a deformable object This category in cludes any tasks where the object being used is not a rigid body Examples include wiping up a liquid spill wringing out a towel and twisting wires Tool assisted tasks These are tasks that require a one handed tool to complete Examples include using a ham mer screwdriver crowbar or wrench These tasks are often easier to simulate with haptic devices as the shaft of the haptics device that the user grasps naturally simulates the grasping of the tool shaft Multiple finger tasks These require more than one finger to perform Examples include grasping any object pulling a pin by its head or turning a bolt or dial Note that some of these tasks can be simulated in the fine motor con trol category above However that category does not neces sarily take into account the interaction between the fingers 3 1 Forces and Torques We categorized force and torque requirements for haptic simulation as follows e Force Only I is a force direction aligned with the mo tion such as pushing a door e Force Only II is a force direction not aligned with the motion such as sanding with a block of wood e Torque Only Iis a torque axis through grip space such as using a screwdriver e Torque Only II is a torque axis that is not
14. instructions for the removal of the fuel tank 13 The simulation procedure was the same for all three de vices The tools operate the same way for all three plat forms but their appearance was dictated by the particulars of each device Figure shows an image from the start of a simulation for the SpaceMouse A screwdriver is on the upper left and a wrench is on the upper right The simulation con sisted of 9 consecutive actions The first two actions were the removal of the two screws holding the access panel shut After the screws were removed the access panel opened au tomatically The next action for the subject to perform was colored red in order to indicate to the subjects which action was to be done when This was done because the purpose of this research was to evaluate the different devices and not to test the subject s memory of remembering the order of the actions Figure 2 shows an image from a CyberGlove simulation after the access panel was opened The third task colored red was the removal of the bolt on the lower left side of the tank in an actual F 16 this would drain the tank The next three actions dealt with removing the three tubes that _ amp COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE Figure 1 The initial view of the simulation for the SpaceMouse Figure 2 Partially completed simulation for the CyberGlove
15. intainer e g insufficient strength C The task is physically impossible due to human limita tions e g size of any maintainer D The task is physically impossible due to physical limi tations part is just inaccessible or not extractable To make the analysis tractable we elected to study only Cases 2 and 3 with a known performable task sequence A Testing for necessary strength or planning complex maneu vers around obstacles is beyond the scope of this inquiry We accommodated tool use because it is a critical part of most maintenance actions In our study we used three commercially available inter active devices The first was the Phantom manufactured by SensAble Technologies which is a device consisting of a rod attached to a robotic arm that is capable of providing force feedback The second device was a CyberGlove man ufactured by Immersion Corporation that can detect the ex act position of the fingers Although we have the Cyber Grasp force feedback option for the CyberGlove we did not use it as the forces it could apply would only affect finger movements and fine motor control was not the experimen tal issue in our study The third device was a SpaceMouse manufactured by 3D Connexion which is a hand held 6 degree of freedom mouse like device The simulations ran on a high end PC Pentium 4 2 GHz workstation with an nVidia 3D graphics board This paper is organized as follows First we describe background
16. ks braced two handed tasks manipulating a deformable object tool assisted tasks and multiple finger tasks We characterize each of these below Fine motor control These are actions that require very fine finger movements Examples include pushing a pin turning a bolt or inserting a bayonet connector Not all haptic devices have the positional accuracy to properly sim ulate these tasks While it is relatively easy to place a motion tracker on various body parts to know where it is LA bayonet connector operates like the cap of a medicine bottle you must first apply force to push the cap in rotate it while the force is being applied then release the force to remove the cap IEEE P COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE positioned the position of the fingers requires specialized equipment Significant arm strength These tasks include any ac tion that requires a significant amount of arm strength to perform Examples include pulling open a stuck access panel sanding where considerable pressure must be placed on the sanding block or pushing open a heavy door Due to device limitations on the haptic devices these tasks can often not be realistically simulated The best most haptic devices can do is to exert a scaled down force so that the user understands that a significant amount of force is re quired while actually exerting less force Tactile
17. lve 14 S T Venkataraman and T Iberall Dextrous Robot Hands Springer Verlag New York 1988 2 4 bad E IEEE D COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE
18. material on action taxonomy and how that re lates to haptic actions Then we give a physics based ac tion taxonomy and use this to select representative tasks for visual and haptic simulation The experiment environment and protocol are given followed by the experimental results and statistical analysis We conclude with a discussion and suggestions for future work 2 Related Work Venkataraman and Iberall 14 provide a summary of grasps of the human hand One of the earliest grasp tax onomies was summarized by Taylor and Schwarz 12 from a taxonomy originally defined by Schlesinger 10 They describe six different grasps of the hand cylindrical fin gertip hook palmar spherical and lateral Cutkosky and Wright 5 6 defined a hierarchical tree structure taxonomy for grasps This tree while complete for maintenance tasks is not exhaustive of all possible hand grasps for example holding a cigarette between the index and middle fingers is not provided by this model The focus of both of these taxonomies is on how a hand grasps an object and not nec essarily on the action such a grasp may enable or require Srinivasan and Basdogan 11 discuss haptic taxonomies but their article is focused more on categorizing current hap tics research There are many other taxonomies that have been presented in research due to space limitations we can not reference them all There has been a lot of research in the area of comparing compu
19. nd simulation if ques tions arose For consistency across subjects experimenter input consisted of repetition of relevant text from the video Following all six simulation runs they filled out a post experiment questionnaire in which they rated how easy or difficult it was to perform the simulation using the particu lar device to which they were assigned 4 1 Accuracy In order for a participant to manipulate an object such as a bolt with a tool such as a hex screwdriver the tip of the tool had to be moved to within a certain distance of the object This distance is represented by a sphere surround ing the object and is shown to the user as a red translucent sphere as shown in figures 1 and 2 As an aid to the user _ COMPUTER SOCIETY Proceedings of the IEEE Virtual Reality 2003 VR 03 1087 8270 03 17 00 2003 IEEE the sphere would turn yellow when the tool tip was within the sphere and green when the object s manipulation was completed The size of these spheres determined the level of accu racy for the simulation A larger sphere would make it eas ier to complete the task yet lower realism the screwdriver could be far away from the bolt and still manipulate it The sizes of the spheres were determined by informal experi ments performed during development The Phantom had the highest precision and had the smallest sphere radius The CyberGlove had the lowest precision and thus had the largest sphere
20. ntal Results 39 subjects participated in the study average age 23 1 sd 4 9 19 female 20 male There were 13 subjects arbi trarily assigned to each device Groups did not differ statis tically in terms of age or gender nor in the amount of time spent using a computer mouse during an average week Analysis of the average speed across the last 4 simu lations showed that performance speed was significantly lower i e faster for the Phantom than for the other two de vices F 2 36 5 699 p 007 Average performance speeds were 58 0 89 3 and 96 8 seconds for the Phantom SpaceMouse and CyberGlove respectively These results are shown in table 2 Device Average Standard Deviation seconds seconds SpaceMouse Phantom CyberGlove Table 2 Mean Performance Speed per Device Post hoc analyses of the time taken confirmed that sig nificant differences existed between the Phantom group and the other groups but not between the CyberGlove and SpaceMouse groups These results are shown in table 3 Groups Mean Standard Significance OO oiterce Bror SM PH 31 29 12 20 0 038 bid PH 31 29 12 20 0 038 hial CG S 7 54 12 20 0 811 PH 38 83 12 20 0 008 Table 3 Results of Tukey Post Hoc Analysis An additional ANOVA showed that ratings on perceived ease of use of the device did not differ significantly among groups 7 Conclusions Results show that the Phantom users learned to accom plish the simul
21. reports of a device s ease of use 1 Introduction One particular application area that seems to be a natural testbed for haptic interaction is in validating disassembly or maintenance instructions Since instructions are authored for human maintainers reducing overall difficulty avoiding errors and improving safety should reduce costs and im prove productivity The prospect of using haptic simulation to aid in validating maintenance tasks led us to an analysis of possible tasks and ways to provide credible virtual task experiences Maintenance instructions provide a concrete application with verifiable results for haptic analysis Real physical sys tems must be disassembled and repaired by maintenance technicians Design decisions that impact maintainability can incur major cost over the lifetime of a complex system such as a modern aircraft Validation means establishing that a given maintenance task could indeed be performed Validating maintenance tasks in a virtual environment may help designers to create more maintainable systems and can also serve to train technicians without taking an aircraft off the flight line or subjecting its components to the extra wear and tear of task practice In a technology investment agreement titled Service Manuals Generation SMG the Air Force Research Labo ratory Deployment and Sustainment Division AFRL HES and GE Global Research are developing an automated tech nical manual development
22. ter input devices Card et al 4 evaluated a mouse keyboard and joystick for text selection on a CRT Bur dea 3 and Hayward and Astley 7 compare multiple hap tic devices as well as provide criteria for comparing them These comparisons are based on measurements of key fea tures DOFs motion range peak force etc and not exper imental results Payette et al 9 describe the experimental testing of the Pantograph a custom haptic device versus a trackball in zero gravity and regular gravity and presents experimental results in that article Significant research has been done in the area of vir tual reality assembly and disassembly environments in par ticular VADE Virtual Assembly Design Environment by Jayaram et al 8 Our research is builds upon previ ous research done at the University of Pennsylvania 1 2 which also focused on disassembly environments for air craft maintenance 3 Haptic Action Taxonomy The range of actions we describe is not meant to be an exhaustive list of all possible human actions Instead it is a general classification of manual actions to be simulated in a virtual environment with haptics These actions are mostly arm and hand actions as that is what the majority of current haptic research as well as current available haptic devices focuses on The classification consists of actions requiring fine motor control significant arm strength tactile friction cooperative two handed tas
Download Pdf Manuals
Related Search