Safe Reinforcement Learning for Robot Manipulation

In recent years, reinforcement learning has shown great promise in solving sophisticated decision-making problems and has also demonstrated impressive results in robotic settings since it provides robots with novel skills without tedious modeling from human engineers. The objective of this research project is to develop a pipeline that autonomously generates a control policy for a given task through reinforcement learning. This pipeline should feature: (1) a reinforcement learning strategy that learns a control policy while guaranteeing the stability of the learnt policy, (2) efficient learning strategies such as curriculum learning which automatically increments the complexity in simulation, and (3) the rapid deployment and integration of new robot skills.

Faculty Supervisor:

Hsiu-Chin Lin;Gregory Dudek;David Meger;Jean-Philippe Roberge

Student:

Partner:

Sycodal

Discipline:

Computer science

Sector:

Manufacturing; Professional, scientific and technical services

University:

McGill University

Program: