A collection of demos showcasing world models, embodied agents, and recursive co-improvement across virtual and physical environments.
An agent trained through RCI navigates complex game environments, solves long-horizon tasks, and adapts to unseen scenarios.
A robot dog executing locomotion and manipulation skills learned primarily in simulation, with minimal real-world fine-tuning.
Multi-finger robotic hands perform pick-and-place, tool use, and assembly tasks guided by a vision-language-action policy.
Generate editable, physically consistent 3D worlds from text, images, or layouts — then deploy agents inside them instantly.
Multiple embodied agents coordinate in a shared generated world to complete cooperative long-horizon missions.
Policies trained entirely in RCI-generated virtual environments execute successfully on real robots with no additional tuning.
We are continuously generating new demos across game worlds, simulation environments, and real robot platforms. Check back soon or subscribe for updates.