Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning

TLDR

Meta‑reinforcement learning can accelerate skill acquisition by leveraging prior experience, but existing research focuses on narrow task distributions that limit generalization to new behaviors. This work introduces a benchmark of 50 robotic manipulation tasks to evaluate and encourage algorithms that generalize and accelerate learning of entirely new, held‑out behaviors. The authors evaluate seven state‑of‑the‑art meta‑RL and multi‑task learning algorithms on the benchmark. The benchmark shows that while each task can be learned successfully, the algorithms struggle to learn multiple tasks concurrently, even with only ten training tasks, highlighting the need for improved generalization.

Abstract

Meta-reinforcement learning algorithms can enable robots to acquire new skills much more quickly, by leveraging prior experience to learn how to learn. However, much of the current research on meta-reinforcement learning focuses on task distributions that are very narrow. For example, a commonly used meta-reinforcement learning benchmark uses different running velocities for a simulated robot as different tasks. When policies are meta-trained on such narrow task distributions, they cannot possibly generalize to more quickly acquire entirely new tasks. Therefore, if the aim of these methods is to enable faster acquisition of entirely new behaviors, we must evaluate them on task distributions that are sufficiently broad to enable generalization to new behaviors. In this paper, we propose an open-source simulated benchmark for meta-reinforcement learning and multi-task learning consisting of 50 distinct robotic manipulation tasks. Our aim is to make it possible to develop algorithms that generalize to accelerate the acquisition of entirely new, held-out tasks. We evaluate 7 state-of-the-art meta-reinforcement learning and multi-task learning algorithms on these tasks. Surprisingly, while each task and its variations (e.g., with different object positions) can be learned with reasonable success, these algorithms struggle to learn with multiple tasks at the same time, even with as few as ten distinct training tasks. Our analysis and open-source environments pave the way for future research in multi-task learning and meta-learning that can enable meaningful generalization, thereby unlocking the full potential of these methods.