Robot policy training

An AI application domain focused on learning policy models that map observations (image, sensor state, joint vectors) to robot control actions. Training is performed via Imitation Learning, Behavior Cloning, Reinforcement Learning, or their combinations on teleoperation data (e.g., Open-X-Embodiment) or in a loop with a physics or generative simulator (world model). Central to robotic foundation models such as RT-2, UnifoLM-WMA-0, and pi0.