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.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2507.18809 [cs.LG] |
| (or arXiv:2507.18809v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2507.18809 arXiv-issued DOI via DataCite |
Submission history
From: Marco Bagatella [view email]
[v1]
Thu, 24 Jul 2025 21:11:39 UTC (939 KB)
[v2]
Wed, 13 May 2026 07:56:01 UTC (2,594 KB)