Intelligent Robot Solutions to Open-World Problems

Abstract

Quals talk:

Robots are increasingly capable of performing tasks in the real world, but much of their use is still confined to heavily controlled environments. To break robots out of the lab and warehouse, they need to be capable of identifying and solving problems autonomously, intelligently, and normatively. In this talk, I will present some of my recent research in creating robot agents that work towards this goal. I will present a system which enables a robot to hypothesize and explore failure cases, a system which enables a robot to construct ``mental simulations'' for symbolic plan repair, and systems which use natural language to adapt to human preferences when correcting behavior. I will conclude by discussing open challenges and ongoing future work.

Research area: Human-Robot Interaction