Skip to content
Modelwire
Subscribe

Replanning framework extends vision-language models to long-horizon robot tasks

Source published ·Modelwire updated

Original coverage: arXiv cs.LG ↗·How Modelwire adds context

The development

Researchers propose RePlan-VLA, a framework that decouples long-horizon task planning from short-horizon robotic execution by layering adaptive replanning on top of existing vision-language-action models. Rather than expecting VLAs to handle multi-step reasoning and error recovery end-to-end, the system monitors real-time progress and revises plans dynamically based on observed outcomes. This approach preserves model simplicity while extending practical deployment horizons, addressing a core limitation in embodied AI where accumulated errors compound across extended manipulation sequences. The work signals growing recognition that scaling robotic capabilities may require orchestration layers above foundation models rather than monolithic end-to-end training.

Modelwire’s AI-generated summary of coverage from arXiv cs.LG.

Modelwire analysis

Explainer

Our AI-generated reading of the wider context and the next developments to watch.

RePlan-VLA doesn't propose a new model architecture or training method, but rather a wrapper that treats existing VLAs as stateless executors. The novelty is in the monitoring layer that detects plan failure and triggers replanning, not in the foundation model itself.

This work directly validates the hybrid decomposition thesis from recent coverage. The 'Isolating Perception for Multi-View Planning' paper (Nov 14) showed that separating perception from reasoning via symbolic solvers recovered 84-97% success rates on constrained tasks. RePlan-VLA applies the same principle to long-horizon robotics, inserting a planning layer above the VLA rather than expecting end-to-end models to handle multi-step error recovery. Similarly, the 'CarryOn' work (Nov 13) tackled context management for long-horizon agents by learning what to retain; RePlan-VLA solves the robotic equivalent by monitoring progress and selectively revising plans. Both papers converge on the insight that monolithic models struggle with extended sequences and benefit from orchestration layers that manage state across multiple steps.

If RePlan-VLA's success rates on standard manipulation benchmarks (e.g., CALVIN, RLBench) exceed the base VLA by more than 15 percentage points without retraining the underlying model, it confirms that planning orchestration is a reliable path to scaling. If the replanning overhead (latency per decision cycle) exceeds 200ms on real hardware, the approach loses practical viability for time-sensitive tasks like assembly.

This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error

Coverage behind this analysis

These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.

  1. ·arXiv cs.CL

    Hybrid planning beats end-to-end VLMs on occluded vision tasks

    A new benchmark exposes a critical failure mode in vision-language models: they collapse under occlusion and partial observability, even on constrained planning tasks. Researchers propose decomposing VLM workflows by isolating perception from reasoning, delegating cross-view synthesis and plan generation to symbolic solvers. The result is dramatic: success rates jump from near-zero to 84-97% across multiple…

    Read Modelwire coverage →Original source ↗

MentionsRePlan-VLA · Vision-Language-Action models · VLA

MW

How this coverage is produced

Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.

Modelwire summarizes, we don’t republish. arXiv cs.LG originally reported this story as “RePlan-VLA: Progress-Aware Replanning for Long-Horizon Vision-Language-Action Models”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.

More to explore

Researchers map the planning mechanism inside vision-language models

arXiv cs.CL·

Retrieval-augmented VLAs overcome domain shift in robot manipulation

Hybrid planning beats end-to-end VLMs on occluded vision tasks

arXiv cs.CL·
Replanning framework extends vision-language models to long-horizon robot tasks · Modelwire