Skip to content

Robotics · 12 November 2026 · 3 min read

NTT says a physics model lifted a stacking success rate to 92.3 percent

NTT said on 11 November that a world model predicts whether a load will collapse or fall before a robot moves. On a small-robot shelf-stacking task the success rate rose from 73.1 percent to 92.3 percent. The company says the first intended use is airport baggage, and that it will run trials with Mujin.

NTT’s diagram of a robot control loop with a world model that predicts the surroundings.
NTT’s diagram of the system, from its 11 November release. It is a diagram. It is not a photograph of a robot or of an airport.

NTT said on 11 November that it has built a world model that, before a robot moves, scores several possible actions for the chance a load will collapse or an object will fall. The company says the model treats weight, orientation and speed as explicit quantities and predicts the next state from equations of motion. It says a component it calls a physics-control adapter is combined with a video-generation model so the system can predict interactions between objects.

On a shelf-stacking task with a small robot, meant to stand in for loading, NTT says the success rate rose from 73.1 percent to 92.3 percent. It says the model also beat earlier methods on a public benchmark, and it does not name that benchmark’s score in the summary of the release. The first application it describes is loading and unloading airport baggage of different shapes and weights. It says it will run field trials with Mujin, and that later uses could include logistics, manufacturing, construction and retail.