ChunkTrust: Adapting Execution Horizons for Robot Policies with Action-Expert Evidence

2026-09-30

AI Summary

This study addresses the limited task adaptability of robotic policies caused by fixed execution step sizes by modeling the step size as a latent variable and proposing a training-free adaptive framework based on dynamic inference from action expert evidence. The core methodology introduces a training-free AHS selector and an optional QHA adapter, which leverage spectral stability and continuity evidence to achieve adaptive planning. To further enhance inference robustness, the framework incorporates Beta posterior tracking, a kernel forgetting mechanism, and query-based attention fusion. Experimental evaluations on the RoboTwin and RoboCasa benchmarks demonstrate that the proposed approach significantly improves task success rates, achieving gains of up to 9.67%. Moreover, real-world deployments confirm its superior performance in practical scenarios.

Abstract

Robot foundation policies predict action chunks, but how many actions to execute before replanning depends on the current task phase. We introduce ChunkTrust, which treats the execution horizon as a latent variable inferred from action-expert evidence rather than a fixed hyperparameter. Its training-free Action-aware Horizon Selector (AHS) combines intra-chunk spectral stability of generation traces with inter-chunk continuity between executed history and predicted actions. An online Beta posterior with kernel forgetting tracks horizon preferences across replans. A lightweight Query-based Horizon Adapter (QHA) optionally learns a context-conditioned dense prior from complementary evidence, fused with current evidence and episode-local Beta memory while the base policy remains frozen. Across RoboTwin2.0 and RoboCasa GR1 Tabletop, AHS improves overall task-averaged success for each evaluated base-policy configuration, including gains of +6.80 percentage points on $π_{0.5}$ over all 50 RoboTwin2.0 tasks and +9.67 percentage points on Qwen3GR00T in RoboCasa. AHS+QHA raises the gain over Base to +9.44 percentage points on the eight-task $π_{0.5}$ evaluation. On four real-world household tasks, AHS improves the equal-task mean normalized process score from 50.4% to 57.5%. Ablations examine the contributions of both evidence terms, temporal memory, and the learned prior. Project page is https://hf618.github.io/ChunkTrust.github.io/

Problem

• robot foundation policies
• execution horizon
• action chunks
• replanning

Innovation

• Action-aware Horizon Selector
• Execution Horizon
• Robot Foundation Policies
• Beta Posterior
• Query-based Horizon Adapter