Greg Stein, George Mason University - Talk Abstract

Abstract:

The next generation of service and assistive robots will need to operate under uncertainty, expected to complete tasks and perform well despite missing information about the state of the world or over their future. Many emerging approaches turn to learning to overcome the challenges of planning under uncertainty, yet can be brittle and myopic, limiting their effectiveness. Our work introduces a family of model-based approaches to long-horizon planning under uncertainty that augments planning with estimates from learning, allowing both high-performance and reliability.

I will present several ongoing projects aimed at improving long-horizon planning in uncertain environments for both single-robots and multi-robot teams. Our learning-augmented planning abstractions afford introspection via counterfactual reasoning, allowing fast and reliable deployment-time policy selection and thus improved performance even in unfamiliar environments dissimilar from any seen during training. The talk will culminate in a discussion of our lab’s ongoing efforts to unite these contributions under a unified planning framework.