DARLA: Improving Zero-Shot Transfer in Reinforcement Learning

TL;DR

DARLA introduces a multi-stage reinforcement learning agent that learns disentangled visual representations to enable robust zero-shot transfer across diverse domains without target domain data.

Contribution

It proposes DARLA, a novel RL agent that learns to see before acting, improving zero-shot transfer by disentangling environment representations.

Findings

DARLA outperforms baselines in zero-shot domain adaptation.

Effective across multiple RL environments and algorithms.

Disentangled representations enhance robustness to domain shifts.

Abstract

Domain adaptation is an important open problem in deep reinforcement learning (RL). In many scenarios of interest data is hard to obtain, so agents may learn a source policy in a setting where data is readily available, with the hope that it generalises well to the target domain. We propose a new multi-stage RL agent, DARLA (DisentAngled Representation Learning Agent), which learns to see before learning to act. DARLA's vision is based on learning a disentangled representation of the observed environment. Once DARLA can see, it is able to acquire source policies that are robust to many domain shifts - even with no access to the target domain. DARLA significantly outperforms conventional baselines in zero-shot domain adaptation scenarios, an effect that holds across a variety of RL environments (Jaco arm, DeepMind Lab) and base RL algorithms (DQN, A3C and EC).