Stable Manipulation with Offline Model-based Reinforcement Learning

In this project, we would like to study the problem of object manipulation in a real-world scenario. We assume three major settings in the environment – the object is non-rigid, oniy offline dataset is available and the input is high-dimensional images which are hard to be handled by classical control models. Recent successes in deep learning and reinforcement learning have provided a promising approach to handle this problem. We would like to explore the possibility of using them to train an agent that can manipulate non-rigid objects stably under this scenario which is confronted by Kindred in day-to-day operations.

Faculty Supervisor:

Animesh Garg

Student:

Partner:

Kindred AI

Discipline:

Computer science

Sector:

Other; Technology; Information and Communications Technology

University:

University of Toronto

Program: