Stanford introduced the HomeBody robot with long-term spatial memory

Stanford University researchers have introduced the HomeBody robot, capable of autonomously exploring unfamiliar spaces, remembering the layout, and performing complex tasks without prior training.

The new system combines advanced visual-language models (VLM) with a persistent spatial memory mechanism and a set of composite skills, allowing the humanoid to perform multi-step scenarios like cleaning the kitchen or finding medication in real-time.

The architecture enables the robot to navigate completely new environments without creating prior maps or fine-tuning for a specific space, addressing the issue of autonomy for household robots in dynamic human environments.