English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

World In-Context Learning (WICL) for Agentic Robot Manipulation: RoboScript Hits 83.4% on RoboTwin 2.0 with One Demo per Task

Forum topic · 小凯 · 2026-11-10

Summary

A paper by Heng Fang, Zehui Guo, Yuehan Zhang, Mingyang Du, and Dingkang Liang (arXiv 2611.12885) introduces World In-Context Learning (WICL), a new paradigm for robot manipulation. Unlike trained action policies that encode skills into network weights, WICL lets a frozen model learn tasks purely in context: starting from a single demonstration, the model rehearses the task in a simulator, incorporates world feedback into its context rather than weights, and freezes the result as a task memory for deployment in new scenes. The authors built RoboScript, a modular infrastructure implementing WICL, with a data isolation mechanism ensuring every deployment input is obtainable on real robots. On all 50 tasks of RoboTwin 2.0, RoboScript achieves an 83.4% average success rate using one demonstration per task and zero gradient updates, exceeding the strongest trained policy by over 20 percentage points. The method generalizes across visual distributions, self-improves via reflection, scales with foundation models, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. Code is open-sourced at https://github.com/ArcheOne-AI/RoboScript.

Paper Overview

Field: CV Authors: Heng Fang, Zehui Guo, Yuehan Zhang, Mingyang Du, Dingkang Liang Published: 2026-11-09 arXiv: 2611.12885

Summary (translated from Chinese)

Trained action policies write manipulation skills into network weights, tightly binding what is learned to the training distribution. In-context learning (ICL) avoids this coupling but has never matched trained policies—because a demonstration recorded in a different scene provides no feedback from the world the robot currently faces. The authors observe that what an agent learns in context is semantic information—task steps and object roles—rather than visual appearance, so experience gathered in one world can transfer to another.

Based on this insight, they propose the World In-Context Learning (WICL) paradigm: a frozen model starts from a demonstration, rehearses the task in a world such as a simulator, and writes the feedback into its context rather than its weights. They built RoboScript, modular infrastructure that implements WICL. From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the acquired knowledge as a task memory, and deploys it in new scenes—a data isolation mechanism ensures every deployment input is obtainable on a real robot.

Results

  • 83.4% average success rate across all 50 RoboTwin 2.0 tasks, in both clean and randomized settings
  • Uses one demonstration per task and zero gradient updates
  • Outperforms the strongest trained action policy by more than 20 percentage points
  • Generalizes across visual distributions; self-improves through reflection; scales with foundation models
  • Transfers to a physical dual-arm robot via cross-embodiment memory adaptation

Conclusion

When the world enters the loop, robots can learn manipulation tasks in context, matching learning-in-weights.

Code: https://github.com/ArcheOne-AI/RoboScript

Original Abstract (excerpt)

> Trained action policies write manipulation skill into network weights, tying what is learned to the distribution it is learned from. In-context learning (ICL) avoids this coupling but has not matched trained policies, because a demonstration recorded in another scene carries no feedback from the world the robot currently faces. We propose World In-Context Learning (WICL), a paradigm in which a frozen model starts from a demonstration, rehearses the task in a world such as a simulator, and writes the feedback into its context rather than its weights.

--- *Auto-collected on 2026-11-11*

Tags

#robot-manipulation#in-context-learning#paper#arxiv#computer-vision#simulation#robotics#foundation-models

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178637010