Continual reinforcement learning deployed in real-life using policy distillation and sim2real transfer

Abstract

We focus on the problem of teaching a robot to solve tasks presented sequentially, i.e., in a continual learning scenario. The robot should be able to solve all tasks it has encountered, without forgetting past tasks. We provide preliminary work on applying Reinforcement Learning to such setting, on 2D navigation tasks for a 3 wheel omni-directional robot. Our approach takes advantage of state representation learning and policy distillation. Policies are trained using learned features as input, rather than raw observations, allowing better sample efficiency. Policy distillation is used to combine multiple policies into a single one that solves all encountered tasks.

Arxiv PDF

Bibtex

@inproceedings{Kalifou19,
  title={Continual Reinforcement Learning deployed in Real-life using PolicyDistillation and Sim2Real Transfer},
  author={Kalifou, René Traoré and Caselles-Dupré, Hugo and Lesort, Timothée and Sun, Te and Diaz-Rodriguez, Natalia and Filliat, David },
  booktitle={ICML Workshop on Multi-Task and Lifelong Learning},
  year={2019}
}