Decoupling Return-to-Go for Efficient Decision Transformer

TL;DR

This paper introduces Decoupled Decision Transformer (DDT), which simplifies the original DT architecture by using only the latest RTG for action prediction, leading to improved performance and efficiency in offline reinforcement learning.

Contribution

The paper identifies a redundancy in DT's use of RTG sequences and proposes DDT, a streamlined model that enhances performance and reduces computational costs.

Findings

DDT outperforms original DT in multiple offline RL tasks.

Using only the latest RTG improves decision transformer performance.

DDT achieves competitive results against state-of-the-art DT variants.

Abstract

The Decision Transformer (DT) has established a powerful sequence modeling approach to offline reinforcement learning. It conditions its action predictions on Return-to-Go (RTG), using it both to distinguish trajectory quality during training and to guide action generation at inference. In this work, we identify a critical redundancy in this design: feeding the entire sequence of RTGs into the Transformer is theoretically unnecessary, as only the most recent RTG affects action prediction. We show that this redundancy can impair DT's performance through experiments. To resolve this, we propose the Decoupled DT (DDT). DDT simplifies the architecture by processing only observation and action sequences through the Transformer, using the latest RTG to guide the action prediction. This streamlined approach not only improves performance but also reduces computational cost. Our experiments show that DDT significantly outperforms DT and establishes competitive performance against state-of-the-art DT variants across multiple offline RL tasks.