Dense Dynamics-Aware Reward Synthesis: Integrating Prior Experience with Demonstrations

TL;DR

This paper presents a reward-shaping framework that combines prior data and expert demonstrations to synthesize dense rewards, significantly improving learning efficiency in sparse-reward reinforcement learning tasks.

Contribution

It introduces a systematic method to generate dense, dynamics-aware rewards by integrating prior experience with demonstrations, reducing manual reward shaping.

Findings

Accelerates learning in sparse-reward RL tasks.

Effectively guides agents to distant goals.

Provides analysis of reward synthesis effectiveness.

Abstract

Many continuous control problems can be formulated as sparse-reward reinforcement learning (RL) tasks. In principle, online RL methods can automatically explore the state space to solve each new task. However, discovering sequences of actions that lead to a non-zero reward becomes exponentially more difficult as the task horizon increases. Manually shaping rewards can accelerate learning for a fixed task, but it is an arduous process that must be repeated for each new environment. We introduce a systematic reward-shaping framework that distills the information contained in 1) a task-agnostic prior data set and 2) a small number of task-specific expert demonstrations, and then uses these priors to synthesize dense dynamics-aware rewards for the given task. This supervision substantially accelerates learning in our experiments, and we provide analysis demonstrating how the approach can effectively guide online learning agents to faraway goals.