Redundancy in Robotics

I think a lot of work needs to be done around testing and evaluation. What assurances do we want before we let the robot go? We don't really have unit tests for robots. And I think that's something that we don't do very much is we don't push our robots to the edge of their performance limits which is what we should do. The other thing that interests me with respect to robustness is deep learning. So you've got a robot, you've trained it and then maybe it's out in the world, it has some new experience. It learns something, right? I think we don't really know as if the robot learns a new thing. Is that on top