NTT Develops Physics-Based World Model to Enhance Robot Action Prediction and Reduce Failures

NTT, Inc. has announced a new physics-based world model that allows robots to predict and evaluate outcomes of candidate actions, such as load collapse or object drops, before execution. This technology, which integrates a proprietary 'physics control adapter' into a video generation model, significantly improved task success rates in a small-robot shelf-loading test from 73.1% to 92.3%. The first envisioned application is automated airport baggage handling.