Backplay: "Man muss immer umkehren"

TL;DR

Backplay is a curriculum learning method for reinforcement learning that uses demonstrations to start training near successful states and gradually moves backwards, improving sample efficiency especially in sparse reward environments.

Contribution

We introduce Backplay, a novel demonstration-based curriculum method that enhances sample efficiency in reinforcement learning by starting near successful states and moving backwards during training.

Findings

Backplay improves training speed in environments with sparse rewards.

It outperforms reward shaping, behavioral cloning, and reverse curriculum methods.

Effective in large grid worlds and complex zero-sum games like Pommerman.

Abstract

Model-free reinforcement learning (RL) requires a large number of trials to learn a good policy, especially in environments with sparse rewards. We explore a method to improve the sample efficiency when we have access to demonstrations. Our approach, Backplay, uses a single demonstration to construct a curriculum for a given task. Rather than starting each training episode in the environment's fixed initial state, we start the agent near the end of the demonstration and move the starting point backwards during the course of training until we reach the initial state. Our contributions are that we analytically characterize the types of environments where Backplay can improve training speed, demonstrate the effectiveness of Backplay both in large grid worlds and a complex four player zero-sum game (Pommerman), and show that Backplay compares favorably to other competitive methods known to improve sample efficiency. This includes reward shaping, behavioral cloning, and reverse curriculum generation.