News · 2026-06-29
A robot AI that adapts to a moved camera by wiggling, not retraining
Researchers have introduced In-Context World Modeling, a technique that lets a robot's AI brain adapt to a changed setup - a moved camera, a different robot body - on the fly, without any retraining. It does this by having the robot perform a few seconds of exploratory fiddling and learning the new configuration from what it observes.
The background: a popular kind of robot AI is the vision-language-action model, which takes in what the robot sees and a description of the task, and outputs the actions to do it. These models are powerful but brittle. Shift the camera to a new angle, swap in a slightly different robot arm, and performance can collapse, because the model was trained on one specific setup and quietly assumes the world still looks exactly like that. The usual fix is to gather new data and retrain or fine-tune the model for each new configuration - slow, expensive, and impractical if you want robots that just work when something changes.
In-Context World Modeling reframes the problem. Instead of treating a new setup as something to retrain for, it treats it as something to figure out in the moment - the way a person handed an unfamiliar tool gives it a few exploratory wiggles to learn how it responds before using it for real. The robot performs a short burst of self-generated, task-agnostic interactions - small movements that aren't about the task, just about probing how this particular system behaves - and the model reads that recent history to infer the essential variables: where the camera is now, how this arm moves, how the world responds to its actions. It builds this understanding inside its context window, the working memory it already uses, and crucially it does so without changing any of its internal weights.
That 'no weight changes' part is what makes it efficient, and it borrows a trick from language models. Big chatbots can learn a new task from a couple of examples you type into the prompt - called in-context learning - without being retrained. In-Context World Modeling ports that idea to physical control: the robot learns the new setup from a few interactions held in context, the same way a chatbot learns a format from a few examples. The analogy: it's the difference between sending an experienced driver back to driving school every time they rent an unfamiliar car, versus letting them adjust the mirrors and feel out the pedals in the parking lot for thirty seconds first.
The reported results show the method significantly outperforming standard vision-language-action baselines when the camera viewpoint is novel, in both simulation and on real robots. That's exactly the kind of everyday change - someone bumped the camera, you mounted it slightly differently - that breaks ordinary policies.
Why it matters: brittleness to setup changes is one of the biggest practical barriers to deploying robots outside carefully controlled labs. A method that adapts from a few seconds of probing, with no retraining, points toward robots that can be moved, reconfigured, or rebuilt without an engineering project each time. It's part of a broader wave of work on world models - AI that understands how environments behave - and a sign that the in-context-learning paradigm that transformed language AI is now reshaping robotics.
The honest caveat is that in-context adaptation has a ceiling set by what the underlying model already implicitly knows. Wiggling to discover a moved camera works because the model has seen many camera angles; a truly alien robot body or a wildly out-of-distribution environment may still demand real retraining, because no amount of probing can teach the model something it has no prior basis to understand. For the common, mundane case of 'same robot, the setup shifted a bit,' though, skipping the retraining step is a genuine and useful win.