Autonomy Talks - James Harrison: Bayesian Models for Adaptive Learning and Control

Autonomy Talks - 22/02/2021

Speaker: James Harrison, Autonomous Systems Lab, Stanford University

Title: Bayesian Models for Adaptive Learning and Control

Abstract: Safe deployment of truly autonomous robots will require that these agents reason about interaction with new, unstructured environments. To ensure good performance and safe operation, a robot must be capable of learning efficiently and characterizing uncertainty due to lack of information. Moreover, this uncertainty must be used to guide safe exploration and exploitation to ensure task satisfaction. In this talk, we introduce a framework for efficient, Bayesian few-shot, adaptive, and open-world learning in both static and changing environments. We then discuss application of these tools to robot modeling, with application to planetary rovers. Finally, we discuss approaches to safe exploration and control under uncertainty using the learned models.

If you want to know more about Autonomy Talks, check https://idsc.ethz.ch/research-frazzoli/autonomy-talks.html

Видео Autonomy Talks - James Harrison: Bayesian Models for Adaptive Learning and Control канала Autonomy Talks