Transferring Spatial Perception Between Robots Operating In A Shared Workspace
1 like668 views
The document presents research on transferring spatial perception between robots in a shared workspace, focusing on the iCub robot's manipulation capabilities. It discusses current localization approaches, a learning framework called Mobee, and proposes future work on unsupervised learning and online learning for spatial representations. Key findings include the use of neural networks for spatial perception and the importance of eye-hand coordination.
Transferring Spatial Perception Between Robots Operating In A Shared Workspace
- 1. Jürgen ’Juxi’ Leitner S. Harding, M. Frank, A. Förster, J. Schmidhuber istituto dalle molle di studi sull’intelligenza artificiale università della svizzera italiana idsia / usi / supsi transferring spatial perception between robots operating in a shared workspace #iros 2012
- 2. our iCub setup is for manipulation
- 3. visual perception Harding et al., GPTP 2012, Leitner et al., ICDL 2012 IROS 2012
- 5. learning approach +
- 7. COLLISIONS ARE BAD
- 8. MoBeE framework Frank et al., ICINCO, 2012.
- 9. MoBeE framework Frank et al., ICINCO, 2012.
- 10. 9DOF bounding box iCub 6 per eye . . . fully connected . . . fully connected Cartesian Coordinates spatial perception neural network
- 11. 600 400 -200 550 300 -300 500 200 -400 450 100 -500 0 -600 400 Y (mm) X (mm) Z (mm) -100 -700 350 -200 -800 300 -300 -900 250 -400 -1000 200 -500 -1100 150 -600 Predicted -1200 Expected 100 -700 -1300 0 200 400 600 800 1000 0 200 400 600 800 1000 0 Sample Index Sample Index -200 -300 -400 -500 -600 Z (mm) -700 -800 -1000 -900 results ANN -1100 -1200 -1300 0 200 400 600 800 1000 Sample Index
- 13. future work unsupervised learning different spatial representations “intrinsic motivation” learning online, continuous learning eye-hand coordination
- 14. thanks for listening juxi@idsia.ch http://Juxi.net/projects video at http://Juxi.net/projects/im-clever/year3.html