Embodied AI, full stack: body, brain, and safety

What the talk covers

• Body — robot morphologies, simulators, active perception, navigation and grasping, tactile sensing, data collection, and sim-to-real transfer.
• Brain — world models: from embedded policies, to RL environments, to video generation, mapped along an evolution timeline and architecture paradigms.
• Safety — the duality of capability and risk, with attack surface and defenses across the perceive → cognize → plan → act stack.
• Built on three authoritative surveys, woven into one panoramic picture of the field.