1. Summary of ARM Research: Results, Issues, Future Work Kate Tsui UMass Lowell January 8, 2006 Kate Tsui UMass Lowell January 8, 2006

2. Summary of Work  Programming the ARM  Accomplishments  Challenges  Future Work  Programming the ARM  Accomplishments  Challenges  Future Work

3. Programming the ARM  Control Modes  Receiving and Sending Packets  Structure of Communication Packet  Control Modes  Receiving and Sending Packets  Structure of Communication Packet

4. Control Modes  Manual and Transparent Control Modes  Both are capable of Joint Movement and Cartesian Movement.  Manual and Transparent Control Modes  Both are capable of Joint Movement and Cartesian Movement.  Joint Movement - One of six joints move at a given time.  Cartesian Movement - Wrist moves linearly in 3D space; joints may move simultaneously.

5. Control Modes: Manual Control  The maximum velocity is 9 cm/s.  Using Cartesian Movement, the ARM can only move linearly in X, Y, or Z.  Math processor handles safety checking, Cartesian coordinate transform checking, and calculation of necessary motor torques for velocity inputs.  The maximum velocity is 9 cm/s.  Using Cartesian Movement, the ARM can only move linearly in X, Y, or Z.  Math processor handles safety checking, Cartesian coordinate transform checking, and calculation of necessary motor torques for velocity inputs.

6. Control Modes: Transparent Mode  The maximum velocity is 25 cm/s.  Using Cartesian Movement, the ARM can simultaneously move in X, Y, and Z.  Math processor is bypassed; safety check is not done.  The maximum velocity is 25 cm/s.  Using Cartesian Movement, the ARM can simultaneously move in X, Y, and Z.  Math processor is bypassed; safety check is not done.

7. Communication: Receiving and Sending Packets  Communication thread is spawned during program initialization.  Based on single producer, single consumer donut factory problem.  Incoming packets from the ARM are stored in reader 10,000 slot semaphore.  Outgoing packets to the ARM are stored in writer 10,000 slot semaphore.  Communication thread is spawned during program initialization.  Based on single producer, single consumer donut factory problem.  Incoming packets from the ARM are stored in reader 10,000 slot semaphore.  Outgoing packets to the ARM are stored in writer 10,000 slot semaphore.

8. Communication: Receiving and Sending Packets  ARM sends packet to PC every 20 ms (hardware limitation).  3 types of incoming packets.  ID = 0x350, 0x360, 0x37F  ARM sends packet to PC every 20 ms (hardware limitation).  3 types of incoming packets.  ID = 0x350, 0x360, 0x37F

9. General Communication Packet

10. Cartesian Mode Packet Interpretation

11. Cartesian Mode Packet Transmission  Velocity: p/(20 * 10 -3 ) mm/s

12. Accomplishments  Design of initial interface  Demo at Robotics: Science and Systems 2006 Conference Workshop on Manipulation for Human Environments  User Testing  Paper presentation (forthcoming) at AAAI-07 Spring Symposium Series Workshop on Multidisciplinary Collaboration for Socially Assistive Robotics  Design of initial interface  Demo at Robotics: Science and Systems 2006 Conference Workshop on Manipulation for Human Environments  User Testing  Paper presentation (forthcoming) at AAAI-07 Spring Symposium Series Workshop on Multidisciplinary Collaboration for Socially Assistive Robotics

13. Interface Design  Interface is compatible with single switch scanning.  Left:  Original image is quartered.  Quadrant containing the desired object is selected.  Middle:  Selection is repeated a second time.  Right:  Desired object is in 1/16th close-up view.

14. Demo

15. User Testing: Hypotheses  H1: Users will prefer a visual interface to a menu based system.  H2: With greater levels of autonomy, less user input is necessary for control.  H3: It should be faster to move to the target in computer control than in manual control.  H1: Users will prefer a visual interface to a menu based system.  H2: With greater levels of autonomy, less user input is necessary for control.  H3: It should be faster to move to the target in computer control than in manual control.

16. User Testing: Experiment  Participants  12 participants (10 male, 2 female)  Age: [18, 52]  67% technologically capable  Computer usage per week (including job related):  67% 20+ hours; 25% 10 to 20 hours; 8% 3 to 10 hours  1/3 had prior robot experience:  1 industry; 2 university course; 1 “toy” robots  Participants  12 participants (10 male, 2 female)  Age: [18, 52]  67% technologically capable  Computer usage per week (including job related):  67% 20+ hours; 25% 10 to 20 hours; 8% 3 to 10 hours  1/3 had prior robot experience:  1 industry; 2 university course; 1 “toy” robots

17. User Testing: Experiment Methodology  Two tested conditions: manual and computer control.  Input device was single switch for both controls.  Each user performed 6 runs (3 manual, 3 computer).  Start control was randomized and alternated.  6 targets were randomly chosen.  Two tested conditions: manual and computer control.  Input device was single switch for both controls.  Each user performed 6 runs (3 manual, 3 computer).  Start control was randomized and alternated.  6 targets were randomly chosen.

18. User Testing: Experiment Methodology  Neither fine control nor depth existed in implementation of computer control during user testing.  In manual control, users were instructed to move the opened gripper “sufficiently close” to the target.  Neither fine control nor depth existed in implementation of computer control during user testing.  In manual control, users were instructed to move the opened gripper “sufficiently close” to the target.

19. User Testing: Experiment Methodology  Manual Control Procedure, using single switch and single switch menu:  Unfold ARM.  Using Cartesian movement, maneuver opened gripper “sufficiently close” to target.  Manual Control Procedure, using single switch and single switch menu:  Unfold ARM.  Using Cartesian movement, maneuver opened gripper “sufficiently close” to target.

20. User Testing: Experiment Methodology  Computer Control Procedure:  Turn on ARM.  Select image using single switch.  Select major quadrant using single switch.  Select minor quadrant using single switch.  Color calibrate using single switch.  Computer Control Procedure:  Turn on ARM.  Select image using single switch.  Select major quadrant using single switch.  Select minor quadrant using single switch.  Color calibrate using single switch.

21. User Testing: Results H1: Users will prefer a visual interface to a menu based system.  83% stated preference for manual control in exit interviews.  Likert scale rate of manual and computer control (1 to 5) showed no significant difference in user experience preference.  H1 was not proven.  Why? Color calibration H1: Users will prefer a visual interface to a menu based system.  83% stated preference for manual control in exit interviews.  Likert scale rate of manual and computer control (1 to 5) showed no significant difference in user experience preference.  H1 was not proven.  Why? Color calibration

22. User Testing: Results H2: With greater levels of autonomy, less user input is necessary for control.  In manual control, counted the number of clicks executed by users during runs, divide by run time. This yields average clicks per second.  In computer control, the number of clicks is fixed.  H2 was confirmed. H2: With greater levels of autonomy, less user input is necessary for control.  In manual control, counted the number of clicks executed by users during runs, divide by run time. This yields average clicks per second.  In computer control, the number of clicks is fixed.  H2 was confirmed.

23. User Testing: Results H3: It should be faster to move to the target in computer control than in manual control.  Distance to time ratio: moving distance X takes Y time.  Under computer control, ARM moved farther in less time.  H3 was confirmed. H3: It should be faster to move to the target in computer control than in manual control.  Distance to time ratio: moving distance X takes Y time.  Under computer control, ARM moved farther in less time.  H3 was confirmed.

24. Challenges  Vision system  Shoulder camera  Gripper camera  Vision system  Shoulder camera  Gripper camera

25. Evolution of UML Vision System: Shoulder Camera

26. Evolution of UML Vision System: Gripper Camera

27. Current UML Vision System  Shoulder (occupant’s view) camera is a Canon VC-C50i Pan-Tilt-Zoom.  Specifications (NTSC):  340,000 pixels  460 horizontal lines, 350 vertical lines  2:1 interlaced  26x digital zoom  Focal length: [3.5, 91.0] mm  Shoulder (occupant’s view) camera is a Canon VC-C50i Pan-Tilt-Zoom.  Specifications (NTSC):  340,000 pixels  460 horizontal lines, 350 vertical lines  2:1 interlaced  26x digital zoom  Focal length: [3.5, 91.0] mm

28. Current UML Vision System  Gripper camera is CCD Snake Camera.  Specifications (NTSC):  1/4” color CCD  510 x 492 pixels  350 vertical lines  2:1 interlaced  Focal length: 3.1 mm  Processor board located 30 cm from CCD.  Gripper camera is CCD Snake Camera.  Specifications (NTSC):  1/4” color CCD  510 x 492 pixels  350 vertical lines  2:1 interlaced  Focal length: 3.1 mm  Processor board located 30 cm from CCD.

29. Gripper Camera Placement  Our choice was to place camera within gripper.  Camera is inline with axis.  Our choice was to place camera within gripper.  Camera is inline with axis.

30. Gripper Camera Concerns  Wired:  Wires impede movement of ARM.  Wireless:  Image quality.  Placement not within gripper:  Not within axis of movement.  Accidental knocking off of camera.  Folded position.  Wired:  Wires impede movement of ARM.  Wireless:  Image quality.  Placement not within gripper:  Not within axis of movement.  Accidental knocking off of camera.  Folded position.

31. Current/Future Work  Integration of ARM with power wheelchair  Depth extraction (image registration, motion filter, optical flow)  Occlusion  User interface  Initial testing at Crotched Mountain  Integration of ARM with power wheelchair  Depth extraction (image registration, motion filter, optical flow)  Occlusion  User interface  Initial testing at Crotched Mountain