Robotics / news brief
NTT's physics-based world model previews robot failures before motion
- Event
- Published

NTT unveiled a world model that adds a physics-control adapter to video generation so a robot can preview candidate actions and screen for drops or load collapse. NTT says a small shelf-loading test improved from 73.1% to 92.3%; trial count, latency, code and weights were not published, and airport-baggage field trials with Mujin are still planned.
Why it matters
A predictive veto layer could make existing robot planners safer around irregular loads, but the evidence is still a company-run lab test rather than a deployment result.
Sources
- NTT Develops Physical AI Technology That Predicts the Future Before Robots Move to Prevent Failures (external site)NTT · primary source ·
- NTT physical AI predicts the future to prevent robot failures (external site)Impress Watch · coverage ·