Directly Forecasting Belief for Reinforcement Learning with Delays

TL;DR

This paper introduces DFBT, a novel belief estimation method for RL with delays that directly forecasts states from observations, reducing errors and improving performance over existing recursive methods.

Contribution

The paper proposes DFBT, a belief forecasting transformer that directly predicts states, significantly reducing compounding errors and enhancing RL performance with delays.

Findings

DFBT reduces prediction errors in RL with delays.

DFBT outperforms SOTA methods on MuJoCo benchmarks.

DFBT improves learning efficiency through multi-step forecasting.

Abstract

Reinforcement learning (RL) with delays is challenging as sensory perceptions lag behind the actual events: the RL agent needs to estimate the real state of its environment based on past observations. State-of-the-art (SOTA) methods typically employ recursive, step-by-step forecasting of states. This can cause the accumulation of compounding errors. To tackle this problem, our novel belief estimation method, named Directly Forecasting Belief Transformer (DFBT), directly forecasts states from observations without incrementally estimating intermediate states step-by-step. We theoretically demonstrate that DFBT greatly reduces compounding errors of existing recursively forecasting methods, yielding stronger performance guarantees. In experiments with D4RL offline datasets, DFBT reduces compounding errors with remarkable prediction accuracy. DFBT's capability to forecast state sequences also facilitates multi-step bootstrapping, thus greatly improving learning efficiency. On the MuJoCo benchmark, our DFBT-based method substantially outperforms SOTA baselines. Code is available at https://github.com/QingyuanWuNothing/DFBT.