Rituraj Kaushik: Data-efficient adaptation in robotics using priors from the simulator

Abstract

In the real world, a robot might face any unanticipated situation, such as damaged motors, unseen terrain conditions, faults in the sensors, and so on. Instead of aborting its mission when such situations occur, a robot could learn to adapt to those situations using reinforcement learning and continue its mission. However, the current reinforcement learning algorithms require prohibitively long interaction time (from several hours to days) to allow a complex physical robot to learn a new skill. In this talk, we present how we can accelerate the learning process on a real physical robot by leveraging prior knowledge derived from a simulator of the robot. More precisely, we use the model-based robot learning together with the priors from a simulator to allow relatively complex robots, such as a hexapod, to adapt to broken legs and fault in the sensors within a minute of interaction and accomplish the task.

Date
February 27, 2020 10:00 — 10:30
Event
SML Seminar
Location
AI Centre, UCL

Bio

Rituraj Kaushik is a 4th year Ph.D. candidate on Machine Learning & Robotics at INRIA-Nancy (Université de Lorraine), France. His research activity is focused on designing model-based robot learning algorithms to allow a robot to adapt to unanticipated situations such as damages within a few minutes of interaction using prior knowledge derived from simulators. He received his master's degree in Electronics Design & Technology in 2016 from Tezpur Univerity, India.
Avatar
Marc Deisenroth
DeepMind Chair in Artificial Intelligence