Course positioning
CS 237B is the advanced follow-up to CS 237A, shifting from mobility to learning new skills and physical interaction. It covers RL and optimal control, contact dynamics, grasping/manipulation, imitation learning, human intent inference, and system architecture validation, implemented on a mobile manipulation platform.
What it covers
- RL and optimal control: MDPs, LQR, policy gradients
- Contact and manipulation: grasping, non-prehensile manipulation, force control
- Imitation learning: behavior cloning, DAgger, inverse RL
- Human intent inference: shared autonomy
- System architecture: perception-planning-control integration, validation
- ROS practice
- Project
Prerequisites
CS 106A; CME 100; CME 106; AA 171/274 (or CS 237A).
Assessment & workload
Assignments plus a project on a mobile manipulation platform. Check the current syllabus for details.
Who it's for
Robot manipulation, embodied AI, and human-robot collaboration.
Study advice
Take CS 237A and CS 234 first. Pair contact dynamics with physics simulation.
Related courses
Official course description
This course teaches advanced principles for endowing mobile autonomous robots with capabilities to autonomously learn new skills and to physically interact with the environment and with humans. It also provides an overview of different robot system architectures. Concepts that will be covered in the course are: Reinforcement Learning and its relationship to optimal control, contact and dynamics models for prehensile and non-prehensile robot manipulation, imitation learning and human intent inference, as well as different system architectures and their verification. Students will earn the theoretical foundations for these concepts and implement them on mobile manipulation platforms. In homeworks, the Robot Operating System (ROS) will be used extensively for demonstrations and hands-on activities. Prerequisites: CS106A or equivalent, CME 100 or equivalent (for linear algebra), CME 106 or equivalent (for probability theory), and AA 171/274.