Learning From Failures: Efficient Reinforcement Learning Control with Episodic Memory

TL;DR

This paper introduces FEMA, a method that uses episodic memory to store failure experiences, improving reinforcement learning efficiency and stability in robot control tasks, especially in contact-rich environments.

Contribution

The paper presents FEMA, a novel episodic memory module that enhances RL training by preventing recurrent failures, leading to significant sample-efficiency gains and successful real-world robot application.

Findings

FEMA improves sample efficiency by 33.11% on MuJoCo tasks.

FEMA enhances RL stability and exploration in contact-rich environments.

Successful deployment of FEMA on a real-world bipedal robot.

Abstract

Reinforcement learning has achieved remarkable success in robot learning. However, under challenging exploration and contact-rich dynamics, early-stage training is frequently dominated by premature terminations such as collisions and falls. As a result, learning is overwhelmed by short-horizon, low-return trajectories, which hinder convergence and limit long-horizon exploration. To alleviate this issue, we propose a technique called Failure Episodic Memory Alert (FEMA). FEMA explicitly stores short-horizon failure experiences through an episodic memory module. During interactions, it retrieves similar failure experiences and prevents the robot from recurrently relapsing into unstable states, guiding the policy toward long-horizon trajectories with greater long-term value. FEMA can be combined easily with model-free reinforcement learning algorithms, and yields a substantial sample-efficiency improvement of 33.11% on MuJoCo tasks across several classical RL algorithms. Furthermore, integrating FEMA into a parallelized PPO training pipeline demonstrates its effectiveness on a real-world bipedal robot task.