Test-time Offline Reinforcement Learning on Goal-related Experience

TL;DR

This paper introduces GC-TTT, a test-time offline reinforcement learning method that adapts policies to current goals using a novel data selection criterion, improving performance in high-dimensional tasks with modest compute costs.

Contribution

It proposes a goal-conditioned test-time training algorithm that fine-tunes policies during evaluation using relevant experience, enhancing performance over standard offline methods.

Findings

GC-TTT improves policy performance across diverse tasks.

Selective data based on relevance enhances fine-tuning effectiveness.

Performance gains are achieved without increasing model size, just by better compute allocation.

Abstract

Foundation models compress a large amount of information in a single, large neural network, which can then be queried for individual tasks. There are strong parallels between this widespread framework and offline goal-conditioned reinforcement learning algorithms: a universal value function is trained on a large number of goals, and the policy is evaluated on a single goal in each test episode. Extensive research in foundation models has shown that performance can be substantially improved through test-time training, specializing the model to the current goal. We find similarly that test-time offline reinforcement learning on experience related to the test goal can lead to substantially better policies at modest compute costs. We propose a novel self-supervised data selection criterion, which selects transitions from an offline dataset according to their relevance to the current state and quality with respect to the evaluation goal. We demonstrate across a wide range of high-dimensional loco-navigation and manipulation tasks that fine-tuning a policy on the selected data for a few gradient steps leads to significant performance gains over standard offline pre-training. Our goal-conditioned test-time training (GC-TTT) algorithm applies this routine in a receding-horizon fashion during evaluation, adapting the policy to the current trajectory as it is being rolled out. Finally, we study compute allocation at inference, demonstrating that, at comparable costs, GC-TTT induces performance gains that are not achievable by scaling model size.