Embodied AI: Perception, Representation, and Action

Course Description

Robots are learning to see, think, and act in the physical world — but building truly intelligent embodied agents requires more than powerful neural networks. This course takes students on a comprehensive journey from the classical foundations of robotics — kinematics, dynamics, control, and planning — to the frontier of learning-based methods including reinforcement learning, imitation learning, generative policies, and vision-language-action models. Organized around three target platforms (manipulators, mobile robots, and humanoids), the course follows a deliberate "Foundations → Learning" philosophy: every classical concept is connected to a modern method that extends or replaces it, giving students the deep understanding needed to advance the state of the art. Through simulation-based assignments, in-depth survey projects, and seminar discussions, students develop both the technical fluency and research maturity to work at the cutting edge of embodied AI.

Prerequisites: Basic knowledge of linear algebra, probability, random processes, optimization, signal processing, and machine learning.

Course Staff

Instructor

Teaching Assistants

Schedule