Legged Robots: Learning, Dynamics, and Control
This course focuses on giving students experience with deploying controllers on a complex real-world (legged) robot, and how to debug a complex system that relies not only on software but the hardware details. While the details/knowledge covered are centered around legged robots, the skills are relevant to any robot platform. You will learn how to model and control (largely with learning-based tools) a dynamically unstable system, and transfer the results from simulation to hardware. The course will include hands-on exercises building simple robots to develop physical intuition, and simple model-based analyses to make a strong connection between physics and the hardware design of the robot, and control or the software design. The core of the course is a team-project using reinforcement learning (RL) to control a legged robot.
Topics
- Assess different legged robot designs and explain their properties and limitations, and the appropriate control approaches for that hardware.
- Implement state-of-the-art controllers for dynamic locomotion, from simulation to hardware.
- Implement quantitative experiments to assess the controller performance, and to systematically debug dynamic behavior in legged robots.
Official roster learning outcomes
Sources
- Course catalogue ↗2026-2027