Knowing When to Stop: Adaptive Action Chunking via Internal Cross-Attention Dynamics in VLAs

• URL: http://arxiv.org/abs/2609.00908v2
• Date: Fri, 04 Sep 2026 08:35:27 GMT
• Title: Knowing When to Stop: Adaptive Action Chunking via Internal Cross-Attention Dynamics in VLAs
• Abstract summary: Action chunking is a standard execution strategy in modern Vision-Language-Action (VLA) frameworks.<n>Short chunks require frequent inference, whereas long chunks can become misaligned with newly observed states.<n>We address this limitation with an adaptive action chunking approach based on internal cross-attention dynamics in the action expert.
• Score: 4.792917095787488
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Action chunking is a standard execution strategy in modern Vision-Language-Action (VLA) frameworks, but fixed execution horizons impose a trade-off between efficiency and accuracy. Short chunks require frequent inference and may cause oscillatory behavior, whereas long chunks can become misaligned with newly observed states. We address this limitation with an adaptive action chunking approach based on internal cross-attention dynamics in the action expert. We observe that, as the prediction horizon extends, action-to-observation cross-attention becomes increasingly dispersed and its entropy rises toward a plateau. This pattern is associated with higher action prediction error and provides an online signal that the current observation offers limited grounding for further open-loop execution. Based on this observation, we introduce a training-free truncation mechanism that detects sustained high-entropy plateaus and dynamically selects the execution horizon during inference. The method uses attention weights already computed by the policy and introduces negligible additional overhead. Evaluations on $π_{0.5}$ and X-VLA across RoboTwin 2.0, LIBERO, and three real-world manipulation tasks show improved average task success over fixed-horizon and adaptive chunking baselines, while preserving efficient closed-loop control. These results show that cross-attention dynamics can provide a practical internal signal for adaptive action execution in VLAs.

Related papers

• FutureRTC: Real-Time Robot Execution with Anticipatory-Conditioned Action Chunking [51.97839311685636]
  FutureRTC is a plug-and-play adaptation framework that predicts execution-time observations and states for asynchronous VLA control without modifying the underlying policy.<n>We show that FutureRTC achieves robustness to inference delays, resulting in smoother trajectories, faster execution, and consistently higher task success rates.
  arXiv  Detail & Related papers  (2026-07-27T05:10:16Z)
• SegDiff: Segmented Trajectory Diffusion for Consistent and Adaptive Robot Manipulation [65.81476183180527]
  We introduce SegDiff, a closed-loop visuomotor policy that integrates the strengths of both paradigms.<n>SegDiff decomposes demonstrations into motion segments between keyposes and learns to predict the continuous trajectory from the current state to the next keypose.<n>SegDiff demonstrates significant performance gains over existing approaches across various simulated and real-world scenarios.
  arXiv  Detail & Related papers  (2026-07-13T02:48:03Z)
• Spatial Attention: Adapting Execution Horizons for Diffusion Policies via Observation Sensitivity [26.591374083905336]
  We adaptively adjust the execution horizon of sampled action chunks, balancing responsiveness and computational efficiency.<n>We show that, under a fixed budget of chunk samplings, the execution horizon that minimizes the cumulative likelihood drop induced by disturbances decreases as Spatial Attention increases.
  arXiv  Detail & Related papers  (2026-07-06T07:22:04Z)
• VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon [40.99993238069073]
  VLA-Corrector is a lightweight corrective inference framework for action-chunked Vision-Language-Action policies.<n>It can be integrated into different VLA models without further retraining the VLA backbone.<n>It mitigates the trade-off imposed by static horizons between execution and policy-call frequency.
  arXiv  Detail & Related papers  (2026-07-02T07:18:53Z)
• Dynamic Execution Horizon Prediction for Chunk-based Robot Policies [29.482987292744568]
  Action chunking has become a standard design in modern robot policies, where the policy predicts a sequence of actions and executes a fixed number of them instead of acting one step at a time.<n>In practice, the execution horizon is typically chosen through empirical tuning and is highly task-dependent.<n>We propose Dynamic Execution Horizon Prediction (DEHP), an effective method that trains a lightweight execution-horizon prediction branch using online reinforcement learning.
  arXiv  Detail & Related papers  (2026-06-09T19:58:31Z)
• FocalPolicy: Frequency-Optimized Chunking and Locally Anchored Flow Matching for Coherent Visuomotor Policy [75.32174422881958]
  FocalPolicy is a visuomotor policy that combines Frequency Locally-d Chunking with Anchored flow matching.<n>We introduce a foresight composite objective that supervises time-domain alignment within the proximal actions.<n>We design locally anchored sampling to target signal propagation consistency efficiency during flow matching training.
  arXiv  Detail & Related papers  (2026-05-15T13:27:55Z)
• MSACT: Multistage Spatial Alignment for Stable Low-Latency Fine Manipulation [4.439585594082787]
  Real-world fine manipulation, particularly in bimanual manipulation, requires low-latency control and stable visual localization.<n>Existing approaches make different trade-offs: action-chunking policies such as ACT enable low-latency execution and data efficiency but rely on dense visual features without explicit spatial consistency.<n>We introduce a multistage spatial attention module that extracts stable 2D attention points and jointly predicts future attention sequences with a temporal alignment loss.
  arXiv  Detail & Related papers  (2026-05-01T07:35:15Z)
• AnchorRefine: Synergy-Manipulation Based on Trajectory Anchor and Residual Refinement for Vision-Language-Action Models [60.04879435087352]
  Vision-language-action (VLA) policies generate actions within a single unified space.<n>We propose AnchorRefine, a hierarchical framework that factorizes VLA action modeling into trajectory anchor and residual refinement.<n> Experiments on LIBERO, CALVIN, and real-robot tasks demonstrate that AnchorRefine consistently improves both regression-based and diffusion-based VLA backbones.
  arXiv  Detail & Related papers  (2026-04-20T04:25:24Z)
• VLA Knows Its Limits [27.43938890325202]
  We show that varying the execution horizon leads to substantial performance deviations.<n>We propose AutoHorizon, the first test-time method that dynamically estimates the execution horizon for each predicted action chunk.
  arXiv  Detail & Related papers  (2026-02-24T23:48:48Z)
• \ extsc{NaVIDA}: Vision-Language Navigation with Inverse Dynamics Augmentation [50.027425808733994]
  textscNaVIDA is a unified VLN framework that couples policy learning with action-grounded visual dynamics and adaptive execution.<n>textscNaVIDA augments training with chunk-based inverse-dynamics supervision to learn causal relationship between visual changes and corresponding actions.<n>Experiments show that textscNaVIDA achieves superior navigation performance compared to state-of-the-art methods with fewer parameters.
  arXiv  Detail & Related papers  (2026-01-26T06:16:17Z)
• Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach [78.4812458793128]
  We propose textbfTACO, a test-time-scaling framework that applies a lightweight pseudo-count estimator as a high-fidelity verifier of action chunks.<n>Our method resembles the classical anti-exploration principle in offline reinforcement learning (RL), and being gradient-free, it incurs significant computational benefits.
  arXiv  Detail & Related papers  (2025-12-02T14:42:54Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.