CAs for Sensing
- We do have a relatively sophisticated vision system for the CABots.
- However, I'm confident this will need radical improvement for
any robotic camera.
- The existing system is somewhat robust to lighting noise, and handles
position invariance, but only
works on a few objects.
- I think we need to explore a learning solution rather than a
programmed (as CABot3) soultion.
- We thought about improving vision for CABot5 including foveation,
and further work on figure ground separation.
- We manage active vision, but if improved, this could be a real
strength.
- We put in a bump sensor for CABot3.
- We could probably work with a range of these sensors relatively
easily.
- We're thinking about speech recogntion and there are some existing
works on echo-location, but this is probably a stretch for now.
- I also think a relatively tight interaction between sensing and
effecting could lead to improved results.
