1. Group 99 Mobile Vision for Autonomous… Navigation and Reconnaissance Jay Silver Kevin Wortman Advised by Bill Ross

2. Group 99 Vision-aided Navigation for Off-road Robotics ACC Testbed development and DARPA FCS-PerceptOR program  Reconnaissance: Remote vision & monitoring, interactive map/video, and 3D modeling  Navigation: Collision avoidance and improved position estimation Reconnaissance and navigation on limited payload platforms (UGV, UAV, etc.) Autonomous systems to support Future Combat Systems (DARPA TTO, PerceptOR)  Vision systems for time-to-goal/bandwidth-constrained collision avoidance in unstructured outdoor environments (example: lightly wooded scenarios)  All-terrain robotic vehicle (Ruggedized, 2 meter/sec)  Digital fisheye video (180 fisheye,12bit, 60fps, Megapixel)  6-DOF IMU (134 Hz, 3 gyros + 3 accelerometers)  Wireless ethernet (802.11 standard)  Onboard compute (2 Pentium IIIs)  Offboard processing (Dual Pentium IV) Robotic Testbed (ACC - 9332)

3. Group 99 Autonomous Navigation

4. Group 99 Navigation  Resolution = f ( polar radius )  Many resolutions to create new perspectives  Path finding with probabilistic obstacle models

5. Group 99 Navigation  Resolution = f ( polar radius )  Many resolutions to create new perspectives  Path finding with probabilistic obstacle models

6. Group 99 Space Variant Resolutions I(x,y) I(r  r  r   r  r Space Variant Res = f ( r ) Standard Fisheye Res = constant } } } } } } } } } } } } Distorted depth/range perception Accurate depth/range perception

7. Group 99 Space Variant Resolutions I(x,y) I(r  r  r   r  r Space Variant Res = f ( r ) Standard Fisheye Res = constant } } } } } } } } } } } } Distorted depth/range perception Accurate depth/range perception

8. Group 99 Space Variant Resolutions

9. Group 99 Space Variant Resolutions Resolution function is the integral of the angular shift w.r.t. eccentricity Approximate with sum and find a least squares best fit polynomial Space Variant Resolution =

10. Group 99 Space Variant Resolutions Equal shifts along the radius of expansion for objects with equal range. Boundary tracking is simplified. Red = High Angular Shift Blue = Low Angular Shift Bird’s Eye View of Angular Shift Inverse Resolution - degrees/pixel Eccentricity - degrees Space-variant resolution function Sensor limit

11. Group 99 Space Variant Resolutions Fixed resolution Space-variant resolution

12. Group 99 Navigation Resolution = f ( polar radius ) Many resolutions to create new perspectives  Many resolutions to create new perspectives  Path finding with probabilistic obstacle models

13. Group 99 Navigation Resolution = f ( polar radius ) Many resolutions to create new perspectives  Many resolutions to create new perspectives  Path finding with probabilistic obstacle models

14. Group 99 Red = high res. Green = med res. (1/2) Blue = low res. (1/4) Multiple Resolutions Red = high res. Green = med res. (1/2) Blue = low res. (1/4)

15. Group 99 Navigation Resolution = f ( polar radius ) Many resolutions to create new perspectives  Path finding with probabilistic obstacle models

16. Group 99 Navigation Resolution = f ( polar radius ) Many resolutions to create new perspectives  Path finding with probabilistic obstacle models

17. Group 99 Path Finding with Probabilistic Obstacle Models Minimum time to goal heavily rewardedDanger Avoidance heavily rewarded

18. Group 99 Ideal Path Finding with Probabilistic Obstacle Models

19. Group 99 Navigation Resolution = f ( polar radius ) Many resolutions to create new perspectives Path finding with probabilistic obstacle models Path finding with probabilistic obstacle models Reconnaissance Next: Access this presentation tomorrow at: www.jaysilver.net/presentations