Course positioning
CS 381 (EE 381) is Shuran Song's advanced course for graduate students who want to do research on large models + robotics. It discusses the difference between embodied AI and internet AI, learning from raw perception to decisions, and continuous adaptation through hardware/software, with an emphasis on paper reading, critical discussion, and projects.
What it covers
- Embodied AI: definitions, differences from internet AI
- Sensorimotor learning: vision, touch, proprioception
- Policy learning: imitation, RL, world models
- Continual adaptation: online learning, domain shift
- Hardware/software co-design: sensors, actuators, morphology
- Paper reading and debate
- Research project
Prerequisites
Deep learning, robotics/RL basics; research-oriented.
Assessment & workload
Paper reading/presentation/debate plus a project. Check the current syllabus for details.
Who it's for
People who want to do embodied AI research, robot learning, or VLA; students preparing for a PhD.
Study advice
Take CS 227A/CS 237B/CS 234 first. The project needs a real robot or a high-fidelity simulation.
Related courses
Official course description
This is an advanced course that will focus on modern machine learning algorithms for autonomous robots as an embodied intelligent agent. It covers advanced topics that center around 1. what is embodied AI and how it differs from internet AI, 2. how embodied agents perceive their environment from raw sensory data and make decisions, and 3. continually adapt to the physical world through both hardware and software improvements. By the end of the course, we hope to prepare you for conducting research in this area, knowing how to formulate the problem, design the algorithm, critically validate the idea through experimental designs and finally clearly present and communicate the findings. Students are expected to read, present, and debate the latest research papers on embodied AI, as well as obtain hands-on experience through the course projects.