1. Path-Following Autonomous Convoy with Multiple Asynchronous Nodes Kyle Lemons, Heather Macfie, Tri Pho, G. M. Ewout van Bekkum PACMAN Georgia Institute of Technology School of Electrical and Computer Engineering Preliminary Design Review ECE 4007, L01 DK2 March 17, 2010

2. Project Overview Proof-of-concept prototype Two or more Autonomous Convoy Vehicles (ACVs) Path-follow algorithm Cost: $300 Military convoy applications Reduction of human requirements Supplement to existing navigation systems Alternative to complex inter- vehicle communications

3. Design Objectives Path-following o Follow distance: 100 cm o Max deviation from path: 10 cm Autonomous operation o Speed: 60 cm/s o Turning radius: 50 cm Passive operation Modular ACVs

4. Project Schedule Completed on or ahead of schedule o Parts ordered o Wheel encoders and infrared cameras mounted, wired, and communicating with FPGA o PWM implemented to control steering and acceleration o FPGA powered and equipped with basic track and follow Ultrasonic range finders may be unnecessary Upcoming work o Custom PCB development begins March 22 o Algorithm determination and optimization begin March 29 o Sensor bar mounting begins April 2

5. System Module Interaction

6. Current Prototype

7. Component Protocols and Standards

11. Robot Vision Options for optically tracking preceding vehicle: o Vision sensor: CMUcam o Laser rangefinder: Neato Robotics’ Revo LDS o Infrared sensor: Nintendo’s Wii Remote Images: http://www.cmucam.org; http://www.hizook.com/blog/2009/12/20/ultra-low-cost-laser-rangefinders- actualized-neato-robotics; http://www.gadgetspage.com/toys-games/how-does-the-wii-remote-work.html CMUcam Revo LDSWii Camera

12. Wii Remote's PixArt Infrared Camera Sensitive to any bright light source o IR pass filter isolates IR wavelengths Fast embedded blob tracking o Up to four blobs at once o Refresh rate of 100 Hz Communication over I²C protocol o Supports fast mode (400 kbit/s) Field of view of 40° o Initially troublesome

13. Resolved Problem: Configuration of Camera and IR LEDs Initially, the sensor bar (the IR LEDs) was to be placed horizontally o The issue arose for discerning the difference between turning car and distant car o Stereoscopic vision for depth perception was the fix

14. Limited Field of View In order to meet design specifications, more than two cameras were needed Stereoscopic vision required an overlap of camera field of views

15. Thinking in the Wrong Direction Attempting to compensate for poor initial configuration Reorient the sensor bar vertically o Resolves turning ambiguity o Simplifies relative location calculation Requires three cameras instead of seven for a 120° FOV

16. Relative Location: Direction

17. Relative Location: Distance

18. PWM Control: Steering and Throttle Pulse Width Modulation (PWM) o Controlled by signal duty cycle o Variable duty cycle o Constant frequency o Simple to implement in hardware RC platform includes PWM control of steering and throttle o Experimentally determined timings o 50 Hz frequency, limited to 5% to 10% duty cycle

19. Steering Control Steering PWM control signal o Calibration required per RC unit o Control module uses 1024 steps Servo motor o Servo motor controller has unknown PWM resolution o Experimentally determined behaviors  Symmetrical "left" and "right" sensitivity  Centered around 7.1%

20. Throttle Control Throttle PWM control signal o Large "idle" dead-band for throttle o Higher "forward" sensitivity than "reverse" o Brake mode function problems Drive motor o Requires higher throttle to "kick-start" movement  Will prove problematic when trying to move slowly o Higher top forward speed than reverse

21. Odometry Where are we on the measured path Estimate our location with odometry o Optical Flow  Optical mouse  Web-cam o Rotary Encoder  Magnetic  Photo-interrupters  Photo-reflector Image: http://www.mccoop.de/images/street-video.png

22. Measure rotation of rear wheels Calculate change in heading and position Implement in the FPGA as a module in VHDL Dead Reckoning with Differential Drive

23. Quadrature Wheel Encoding Pattern inside rear wheels Quadrature encoding o Gray code output o Double resolution o Simpler acceptance testing

24. HLC1395-002 Photo Reflector from Honeywell o Infrared LED and photo-transistor in the same package o 100 mW power dissipation o 0.6 mA photo-transistor on current Inverting Schmitt Trigger o Debounce and convert analog to discrete output Electrical Design of Wheel Encoders

25. Physical Design of Wheel Encoders

26. Problems with Wheel Encoding What if the wheels slips? o Remove power to the rear wheel Maximum resolution? o No optical specifications other than “unfocused” o Testing showed ~36-48 steps per revolution Duty cycle in the gray code exactly 50%? o Photo-reflector measures reflectance, but what if it sees more than one step? o Optimize duty cycle of the light areas on the pattern

27. Path Generation Markers denote lead vehicle position "Visual Snakes" Options o Regression o Cluster averaging Issues for consideration o Computational complexity o Path accuracy

28. Path Traversal Comparing vehicle location with path generated Options o Absolute reference frame o Relative reference frame o Moving reference frame Issues for consideration o Numeric overflow o Pre-/Post-processing o Compounding rounding error Absolute reference Relative reference Moving reference

29. PACMAN Final Prototype o More cameras o Two ACVs o Path-following Current Prototype o One camera o Limited field of view o Basic point-and-steer follow o Coasts to a stop

30. Questions But first! A demo...