NTT announced on 11 November a technology that lets a robot predict the result of several possible actions, such as a load collapsing or an item falling, before it acts. It then picks an action likely to succeed.

It rests on a world model that handles physical quantities such as weight, direction and speed, and predicts changes from equations of motion. NTT built a physical control adapter and added it to a video generation model so it predicts how objects interact, including in scenes absent from its training data.

NTT names airport baggage and small parcels as examples, since their shape, weight and layout vary. The technology can work alongside existing robot control software to support its decisions.

Source: Robotstart

Background

NTT

Telecom and research company

A Japanese company that announced a physical AI technology for robots.