IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures

TLDR

The field faces a key challenge of managing increased data volumes and extended training times. This work aims to solve a large collection of tasks with a single reinforcement learning agent using a single set of parameters. We introduce IMPALA, a distributed actor‑learner architecture that decouples acting and learning, employs V‑trace for off‑policy correction, and scales efficiently to thousands of machines while maintaining data efficiency, demonstrated on DMLab‑30 and Atari‑57. IMPALA achieves superior performance to prior agents with less data and demonstrates positive transfer across tasks through its multi‑task approach.

Abstract

In this work we aim to solve a large collection of tasks using a single reinforcement learning agent with a single set of parameters. A key challenge is to handle the increased amount of data and extended training time. We have developed a new distributed agent IMPALA (Importance Weighted Actor-Learner Architecture) that not only uses resources more efficiently in single-machine training but also scales to thousands of machines without sacrificing data efficiency or resource utilisation. We achieve stable learning at high throughput by combining decoupled acting and learning with a novel off-policy correction method called V-trace. We demonstrate the effectiveness of IMPALA for multi-task reinforcement learning on DMLab-30 (a set of 30 tasks from the DeepMind Lab environment (Beattie et al., 2016)) and Atari-57 (all available Atari games in Arcade Learning Environment (Bellemare et al., 2013a)). Our results show that IMPALA is able to achieve better performance than previous agents with less data, and crucially exhibits positive transfer between tasks as a result of its multi-task approach.