M$^3$PC: Test-time Model Predictive Control for Pretrained Masked Trajectory Model

TL;DR

This paper introduces M$^3$PC, a test-time Model Predictive Control method that enhances pretrained trajectory models for offline RL tasks by using their predictive capabilities during inference, leading to improved decision-making.

Contribution

The paper proposes a novel use of MPC at inference time to leverage pretrained trajectory models as both policy and world models, improving performance without additional training.

Findings

Significant performance improvements on D4RL and RoboMimic benchmarks.

Effective adaptation to Offline to Online RL and Goal Reaching RL scenarios.

Enhanced generalization to different task targets.

Abstract

Recent work in Offline Reinforcement Learning (RL) has shown that a unified Transformer trained under a masked auto-encoding objective can effectively capture the relationships between different modalities (e.g., states, actions, rewards) within given trajectory datasets. However, this information has not been fully exploited during the inference phase, where the agent needs to generate an optimal policy instead of just reconstructing masked components from unmasked ones. Given that a pretrained trajectory model can act as both a Policy Model and a World Model with appropriate mask patterns, we propose using Model Predictive Control (MPC) at test time to leverage the model's own predictive capability to guide its action selection. Empirical results on D4RL and RoboMimic show that our inference-phase MPC significantly improves the decision-making performance of a pretrained trajectory model without any additional parameter training. Furthermore, our framework can be adapted to Offline to Online (O2O) RL and Goal Reaching RL, resulting in more substantial performance gains when an additional online interaction budget is provided, and better generalization capabilities when different task targets are specified. Code is available: https://github.com/wkh923/m3pc.