World In-Context Learning for Agentic Robot Manipulation
World In-Context Learning for Agentic Robot Manipulation
This paper introduces World In-Context Learning (WICL), a framework where a frozen model learns manipulation tasks through simulation rehearsal and context updates instead of modifying network weights. The authors present RoboScript, a modular infrastructure that implements WICL to transfer skills across visual distributions and embodiments. Evaluated on 50 tasks from the RoboTwin 2.0 benchmark, RoboScript achieves an 83.4% mean success rate using only one demonstration per task without gradient updates, outperforming strong trained action policies by over 20 percentage points. The method generalizes across visual domains, improves via reflection, scales with foundation models, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. All code is available at https://github.com/ArcheOne-AI/RoboScript.
The authors propose World In-Context Learning (WICL) and RoboScript, demonstrating that a frozen foundation model can learn manipulation tasks via simulation rehearsal and context updates rather than weight gradients, achieving 83.4% mean success on 50 RoboTwin 2.0 tasks with one demonstration per task.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
improvement over baseline
>20
View original evidence
n and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. With a world in the loop, a robot can learn a manipulation task in its
Open source S5number of tasks
50
View original evidence
modular infrastructure that implements WICL for agentic manipulation. From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the learned knowledge as a Task Memory, and deploys the frozen memory on new scenes with data isolation that ensures every deployment input is available on a real robot. On all 50 tasks of RoboTwin 2.0, RoboScript reaches 83.4% mean success across clea
Open source S4Demonstrations / episodes
1 episodes
View original evidence
n and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. With a world in the loop, a robot can learn a manipulation task in its
Open source S5
- code: Available Artifact link Open source S6
- paper: not_reported
- weights: not_reported
- dataset: not_reported
- code: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
modular infrastructure that implements WICL for agentic manipulation. From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the learned knowledge as a Task Memory, and deploys the frozen memory on new scenes with data isolation that ensures every deployment input is available on a real robot. On all 50 tasks of RoboTwin 2.0, RoboScript reaches 83.4% mean success across clea
Open source S4
n and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. With a world in the loop, a robot can learn a manipulation task in its
Open source S5
context is semantic, recording task steps and object roles rather than visual appearance, so experience gathered in one world transfers to another. Based on this insight, 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. We build RoboScript, a
Open source S3
context as well as it learns one in its weights. All code is available at https://github.com/ArcheOne-AI/RoboScript
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org