Particle Filter Localization for Autonomous AUVs

Senior Thesis Presentation 2013

Transcript

1. Particle Filter Localization for Autonomous AUVs Using Augmented Reality Tags Ed Kelley, 2013 Szymon Rusinkiewicz
2. aka
3. Where is the Quadcopter?
4. Motivation
5. This statue is cool http://www.asergeev.com/pictures/archives/2007/572/jpeg/05.jpg
6. I want a 3d model http://www.asergeev.com/pictures/archives/2007/572/jpeg/05.jpg
7. Video Games Virtual Reality Movies Archeology Architecture Maps Crash Scenes
8. Manual Modeling? Laser Scanner? Multi-View Stereo? Microsoft Kinect?
9. Manual Modeling? Laser Scanner? Multi-View Stereo? Microsoft Kinect?
10. Irschara et al. 2010
11. Easy Cheap Complete High Quality
12. Quadcopters!
13. Related Work
14. Irschara et al.
15. Engel et al.
16. Bills et al.
17. System Design
18. AR.Drone 2.0 http://ardrone2.parrot.com/photos/photo-album/
20. Localization + Controller = Autonomy
21. Localization
22. Local drift. tend to measurements
23. No GPS No rangefinders
25. Kalman Filter? Grid Based Markov? Particle Filter?
26. Kalman Filter? Grid Based Markov? Particle Filter?
27. . This is a particle
28. . It represents a possible pose
29. . x y z heading weight
30. Prediction Step Update the position of each particle using noisy velocity and gyroscope readings.
31. Correction Step 1. Check for an augmented reality tag. 2. Calculate transformation from camera to tag. 3. Use known coordinates of the tag to calculate the position of the quadcopter.
32. Correction Step 4. Weight the particles using their similarity to this calculated position. 5. Perform weighted resampling of the particles. 6. With some probability, replace particles with this calculated position.
33. Estimate Use a linear combination of the particle values to create an estimated pose.
34. Testing
35. Gyroscope
36. Ultrasound
37. AR Tag
38. AR Tag
39. AR Tag
40. AR Tag
41. Manual Flight Test
47. Conclusion
48. Particle filter localization using augmented reality tags performs substantially better than integrated velocity alone.
49. AR Tags are highly dependent on lighting.
50. Its called Hardware for a reason.
51. Next steps... Full integration with controller. Modeling objects.
52. Thanks!