Robust Robot Design, Dependable AI for HRI

Setting: upper limb prosthetics, sensor array, noisy.

Use ML to identify features for user to improve response time and controller. Use AI for control to distinguish between different degrees of freedom. But ML has power requirements, need low-power ML for robot. He does not yet much experience with ML, just basics.



Would like to design the prosthetic himself or the sensors.

Biggest weakness: hasn't done as much with ML/evolutionary algorithms, has some general knowledge of AI, will bring himself up to speed in Year 1 of the CDT.

Skill gaps: bio-inspired computing, intro to ML, hasn't been exposed to what's out there, e.g., which classified to use. Needs to research area to bring him up to speed, but he says he's good at that.

Stefano suggests ML, Adam tries to explain ML.

Barrier: it's new, it's niche, the things we have work well enough, there isn't enough drive.
Who supports if there's a problem? Keep going, keep refining, focus/stress test on actual use cases. Public might be more accepting.