Learning Memory-Dependent Continuous Control from Demonstrations

TL;DR

This paper introduces READER, a novel reinforcement learning algorithm that leverages demonstrations and experience replay to improve memory-dependent continuous control in partially observable environments, enhancing sample efficiency.

Contribution

The paper presents READER, a new algorithm capable of handling memory-dependent control tasks using demonstrations, extending RL to partially observable environments.

Findings

Significantly reduces environment interactions.

Outperforms baseline RL algorithms in sample efficiency.

Effective in memory-critical control tasks.

Abstract

Efficient exploration has presented a long-standing challenge in reinforcement learning, especially when rewards are sparse. A developmental system can overcome this difficulty by learning from both demonstrations and self-exploration. However, existing methods are not applicable to most real-world robotic controlling problems because they assume that environments follow Markov decision processes (MDP); thus, they do not extend to partially observable environments where historical observations are necessary for decision making. This paper builds on the idea of replaying demonstrations for memory-dependent continuous control, by proposing a novel algorithm, Recurrent Actor-Critic with Demonstration and Experience Replay (READER). Experiments involving several memory-crucial continuous control tasks reveal significantly reduce interactions with the environment using our method with a reasonably small number of demonstration samples. The algorithm also shows better sample efficiency and learning capabilities than a baseline reinforcement learning algorithm for memory-based control from demonstrations.