titleHorizon-Scalable Q-Learning
authorsHaoran Xu, Pragya Srivastava, Xinjie Liu, Siddhant Agarwal, Amy Zhang
date2026-11-07
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.10987
affiliationUniversity of Texas at Austin
upvotes1 (Hugging Face Daily Papers, 2026-11-12)
linksabs pdf hf code
cite
relatedAdvances in Agentic Computer Graphics
No Hidden Prompts Needed! You Can Game AI Peer Review with Presentation-Only Revisions
Maglev: Sliding Recurrent Memory
ISO: An RLVR-Native Optimization Stack
CalVerT: Augmenting Agents with Calibrated Verifier Telemetry Improves Action and Learning in Knowledge-Intensive Tasks
abstractLearning accurate off-policy state-action value functions (Q-functions) from offline datasets becomes harder as the task horizon grows. One-step temporal-difference (TD) learning bootstraps after every transition, so small per-step errors accumulate along the horizon, and early states can be badly misvalued even when the TD loss is small. Multi-step TD shortens this chain by following the dataset trajectory for several steps; however, since the behavior policy is often suboptimal, every followed action adds a bias, and no fixed number of steps suits a whole trajectory. In this paper, we propose Horizon-Scalable Q-Learning (HQL), a simple method that enables accurate long-horizon value learning by automatically balancing between one-step and multi-step returns. HQL extends a target through a dataset action only when the critic values it at least as highly as the best-of-N actions sampled from the BC policy, and otherwise falls back to a one-step target. We show that this rule is a principled off-policy correction that needs no action densities, which makes HQL compatible with expressive flow and diffusion policies. Experiments show that HQL avoids the failures of both one-step and multi-step TD on a combination-lock task across horizons, and outperforms prior horizon-reduction methods on long-horizon OGBench and OCBench tasks.

bibtex

@misc{xu2026horizon,
  title={Horizon-Scalable Q-Learning},
  author={Haoran Xu and Pragya Srivastava and Xinjie Liu and Siddhant Agarwal and Amy Zhang},
  year={2026},
  eprint={2611.10987},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.10987}
}