mlsanityData to 7 Oct 2026

Fast Plans, Faithful Actions: Closing the Planning-Execution Gap in Hierarchical Vision-Language-Action Models

arXiv 2609.30833 ↗First posted 25 Sep 2026cs.ROcited by 0

On the map: Vision-language-action models → Efficient VLA inference and adaptation also fits: Reasoning and chain-of-thought in VLA

Abstract (arXiv)

Hierarchical vision-language-action (VLA) systems consist of a high-level vision-language planner and a low-level action expert that generates continuous actions. This hierarchical design has practical value only if the planner can generate plans fast enough to meet real-time control requirements, and the resulting plans actually contribute to the generation of action. We study one such system, a waypoint hierarchy pipeline adapted from π0.5, and find that neither requirement is satisfied. This baseline relies on token-level autoregressive decoding (Token-AR) to generate a waypoint plan, requiring 57 very expensive vision-language model (VLM) forward passes. However, we find that erasing the waypoint endpoints has little effect on task success. Two findings reveal the misalignment of planner-executor: the planner generates outputs at an excessively fine granularity, and the executor underuses plans as a control condition. We address the latency issue with waypoint-aligned block-autoregressive decoding (Block-AR), and plan underuse issue with normalized goal modulation (NGM), a layer-wise goal path constrained by phase gating and anti-shortcut training so that the waypoint influences action generation maintaining other signals. Our method reduces the maximum number of VLM forward passes from 57 to 8 on LIBERO, including one prefix prefill, and achieves an 8.7× reduction in planning latency on a Rokae dual-arm robot. With normalized goal modulation and anti-shortcut training, Block-AR's success rate on LIBERO-Long increases from 91.0% to 96.2%, while its average success rate across the four suites increases from 95.85% to 98.45%. On three bimanual tasks with this robot, success rates remain comparable across methods.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Efficient VLA inference and adaptation, one of 13 directions in Vision-language-action models. Builds on 23 of its 28 references that are in the map. First posted on arXiv 25 Sep 2026.

Builds on

23 of 28 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025CEED-VLA: Consistency Vision-Language-Action Model with Early-Exit Decoding16 Jun
  2. 2025PD-VLA: Accelerating Vision-Language-Action Model Integrated with Action Chunking via Parallel Decoding4 Mar
  3. 2025FAST: Efficient Action Tokenization for Vision-Language-Action Models16 Jan

Other references · 20

  1. 2022Classifier-Free Diffusion Guidancecited by 5,717
  2. 2022Flow Matching for Generative Modelingcited by 5,426
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  4. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  5. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  6. 2017Hindsight Experience Replaycited by 1,096
  7. 2019Causal Confusion in Imitation Learningcited by 286
  8. 2025Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Bettercited by 142
  9. 2025HAMSTER: Hierarchical Action Models For Open-World Robot Manipulationcited by 125
  10. 2024Hierarchical Diffusion Policy for Kinematics-Aware Multi-Task Robotic Manipulationcited by 92
  11. 2020Fighting Copycat Agents in Behavioral Cloning from Observation Historiescited by 70
  12. 2023Waypoint-Based Imitation Learning for Robotic Manipulationcited by 59
8 more
  1. 2025CAST: Counterfactual Labels Improve Instruction Following in Vision-Language-Action Modelscited by 28
  2. 2026LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queriescited by 24
  3. 2024What's the Move? Hybrid Imitation Learning via Salient Pointscited by 9
  4. 2024BiKC: Keypose-Conditioned Consistency Policy for Bimanual Robotic Manipulationcited by 5
  5. 2025NoTVLA: Semantics-Preserving Robot Adaptation via Narrative Action Interfacescited by 4
  6. 2025GAE: Unleashing Physical Potential of VLM with Generalizable Action Expertcited by 4
  7. 2026Coarse-to-Control: Action-Token Planning for Vision-Language-Action Modelscited by 3
  8. 20263D HAMSTER: Bridging Planning and Control in Hierarchical Vision Language Action Models through 3D Trajectory Guidancecited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Depth-Wise Probing and Pruning of the Planning Token in a Driving Vision-Language-Action Model
  2. 2026STeP: Signal Temporal Logic for Precise Specifications for Action Generation with Vision Language Models
  3. 2026Explicit Language Memory for Long-Horizon Planning in Vision-Language-Action Models
  4. 2026Long-Horizon Manipulation via Trace-Conditioned VLA Planning
  5. 2026PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
  6. 2025LoHoVLA: A Unified Vision-Language-Action Model for Long-Horizon Embodied Tasks
  7. 2026Towards Long-horizon Embodied Agents with Tool-Aligned Vision-Language-Action Models
  8. 2026SparkVLA: Stop-Aware Hierarchical VLA with Adaptive Action Chunking for Long-Horizon Manipulation