Academics / Courses / DescriptionsBME 495-0-09: Physical Intelligence and Embodied AI
VIEW ALL COURSE TIMES AND SESSIONS
Description
Recent advances in artificial intelligence have been driven primarily by large-scale learning from static datasets. However, biological intelligence emerges through continuous interaction among brains, bodies, and environments. This course develops physical intelligence—the idea that morphology, mechanics, and the sensorimotor loop perform real computation—and connects it to modern embodied AI, where agents that act in physical or simulated environments learn very differently from disembodied ones. Building on the mechanics, dynamics, controls, and programming background of mechanical engineering students, the course moves from “brainless” mechanical intelligence (passive-dynamic walkers, compliant bodies) through active sensing and feedback to reinforcement learning, planning, and the evolutionary origins of cognition. Students implement embodied agents in simulation and build a final agent that senses, acts, and learns in a predator–prey arena modeled on the instructor’s research.
Learning objectives
By the end of the course, students will be able to:
Articulate the embodiment thesis and explain, with concrete mechanical examples, what it means for a body or an environment to “compute” (morphological computation).
Analyze locomotion and active sensing quantitatively—relating morphology and movement to thrust, sensing volume, and information gain—using tools from dynamics, fluids, and optimization.
Formulate a natural embodied task as a reinforcement-learning problem (MDP, reward, policy, value) and train an agent in a standard simulation framework.
Reason about planning, prediction, and partial observability, and explain why and when planning beats reactive control.
Connect biological and engineered intelligence, including the evolutionary pressures (e.g., sensory range and environmental complexity) that may select for planning.
Design, implement, simulate, and document an embodied agent end-to-end, and communicate the result in writing and a presentation.