Write Once, Plan Many: Building Reusable Visual Planning Systems from One Scene
We present a training-free framework that constructs reusable perception and planning programs from a domain specification and a single practice scene without a paired solution.
ProofPaper ↗
Key points
- Vision-language models (VLMs) can reason about visual planning tasks, but querying them for every instance incurs repeated inference cost and latency.
- Generating reusable planning programs offers an alternative, but constructing reliable planning systems for new problem domains without extensive supervision remains challenging.
- Our framework addresses it by leveraging VLM's ability to provide a renderer and a dynamic model based on its scene understanding.
- Across fifteen domain variants spanning 2D and 3D scenes with discrete and continuous state, our systems achieve 99.3% average planning success, compared with 80.9% for the strongest domain-level program-generation baseline and 90.7% for the strongest instance-level baseline.
Sources (1)
- [1]Write Once, Plan Many: Building Reusable Visual Planning Systems from One ScenearXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 09:52 PM
We present a training-free framework that constructs reusable perception and planning programs from a domain specification and a single practice scene without a paired solution.
Vision-language models (VLMs) can reason about visual planning tasks, but querying them for every instance incurs repeated inference cost and latency.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026Run fast, private agents on-device with LFM2.5-24B-A2B
- Oct 13, 2026Gemini 4 Argon for Robotics
- Oct 13, 2026inclusionAI/Ling-3.1-flash
- Oct 13, 2026Hybrid Reasoning Agents
- Oct 13, 2026Fish Audio Drama 3 Technical Report
- Oct 12, 2026Microsoft’s new AI chip makes it less reliant on Nvidia’s GPUs