This course develops learning-based control through exact and approximate dynamic programming (DP). Approximate DP provides a common framework for interpreting reinforcement learning and connecting learned values, policies, and models to LQR, MPC, and rollout. A central emphasis is online improvement: using lookahead and optimization to improve a learned controller during execution. Modern RL algorithms are then interpreted within this framework rather than treated as isolated recipes.
Course Materials: K. Choi, Learning-Based Control for Mobility Systems, evolving working draft, 2026.
The online version provides the current working draft and may evolve during the course.
For more information, Read Current Working Draft and View Fall 2026 Syllabus (PDF)