microSLAM team says it turns single-camera footage into 3D robot-training environments

The microSLAM team describes a system that maps surroundings and tracks camera position using a single RGB video stream. It says the resulting interactive 3D environments can be used to train behavioral foundation models. The team’s pitch is to capture robot-learning data through head-mounted footage of people doing everyday work, rather than fitting every worker with a camera rig. Its example: a person walks a factory floor, and a robot learns to navigate it.