MIT researchers, in collaboration with the Toyota Research Institute, have developed a new generative AI tool designed to train robots in virtual environments.
The system, called “steerable scene generation”, allows robots to learn from synthetic environments where real-world data may be limited or costly to obtain.
By simulating dynamic and diverse scenes, the tool helps robots develop better adaptability, perception, and interaction skills before being deployed in real settings — significantly reducing training time and resource consumption.
This innovation marks another step forward in bridging AI, robotics, and real-world automation — where generative models are no longer just creating images or text, but entire learning worlds for intelligent machines.
Source: AI Business – MIT releases generative AI tool for virtual robot training