Self-Supervised Policy Adaptation during Deployment

TL;DR

This paper introduces a self-supervised approach enabling reinforcement learning policies to adapt during deployment in new environments without reward signals, significantly improving generalization across diverse simulated and real robotic tasks.

Contribution

It proposes a novel self-supervised training method for policy adaptation during deployment without prior knowledge of environment changes.

Findings

Improves policy generalization in 31 out of 36 environments

Outperforms domain randomization in most cases

Effective in both simulation and real robotic tasks

Abstract

In most real world scenarios, a policy trained by reinforcement learning in one environment needs to be deployed in another, potentially quite different environment. However, generalization across different environments is known to be hard. A natural solution would be to keep training after deployment in the new environment, but this cannot be done if the new environment offers no reward signal. Our work explores the use of self-supervision to allow the policy to continue training after deployment without using any rewards. While previous methods explicitly anticipate changes in the new environment, we assume no prior knowledge of those changes yet still obtain significant improvements. Empirical evaluations are performed on diverse simulation environments from DeepMind Control suite and ViZDoom, as well as real robotic manipulation tasks in continuously changing environments, taking observations from an uncalibrated camera. Our method improves generalization in 31 out of 36 environments across various tasks and outperforms domain randomization on a majority of environments.