When to Re-Commit: Temporal Abstraction Discovery for Long-Horizon Vision-Language Reasoning

TL;DR

This paper introduces a learnable, state-conditioned commitment depth for long-horizon vision-language reasoning, improving efficiency and success rates over fixed-depth methods.

Contribution

It formalizes commitment depth as a dynamic, learnable policy component and demonstrates its effectiveness in vision-language tasks.

Findings

Adaptive policy outperforms fixed-depth baselines by up to 12.5% in solve rate.

Uses 25% fewer primitive actions per episode.

Outperforms GPT-5.5 and Claude Sonnet on tested tasks.

Abstract

Long-horizon reasoning requires deciding not only what actions to take, but how deeply to commit before the next observation. We formalize this as \emph{commitment depth}: the number of primitive actions executed open-loop between replans. Commitment depth induces a trade-off between replanning cost and compounding execution error, yet most existing long-horizon systems fix it as a hand-designed scalar. In this work, we instead treat commitment depth as a learnable, state-conditioned variable of the policy itself. We instantiate this within a model-native vision--language policy that jointly predicts both what to execute and for how long. Across Sliding Puzzle and Sokoban, the resulting adaptive policy Pareto-dominates every non-degenerate fixed-depth baseline, achieving up to 12.5 percentage points higher solve rate while using approximately 25\% fewer primitive actions per episode. Despite using a 7B backbone, our method outperforms GPT-5.5 and Claude Sonnet on both tasks, while every tested open-weight vision--language model achieves 0\% zero-shot success. We further present a theoretical analysis showing that, under the standard commitment-depth surrogate, state-conditioned commitment strictly dominates any fixed depth whenever the locally optimal depth varies across states.