Feedback, classical and optimal control, imitation learning, and reinforcement learning with MuJoCo experiments.
Simeon Nedelchev · 12 lectures · 7 practices · https://simeon-ned.github.io/forc/
Lectures
| Week | Lecture | Materials |
|---|---|---|
| 1 | 1. Feedback, state, and control objectives | Notes · Slides · Practice |
| 1 | 2. Simulation and the MuJoCo control loop | Notes · Slides · Practice |
| 2 | 3. Jacobian control and differential inverse kinematics | Notes · Slides · Practice |
| 2 | 4. PD, PID, and state feedback | Notes · Slides · Practice |
| 3 | 5. Gravity compensation and inverse dynamics | Notes · Slides · Practice |
| 3 | 6. LQR and trajectory optimization | Notes · Slides · Practice |
| 4 | 7. Constrained model predictive control | Notes · Slides · Practice |
| 4 | 8. Sampling-based predictive control and MPPI | Notes · Slides · Practice |
| 5 | 9. Behavior cloning and policy evaluation | Notes · Slides · Practice |
| 5 | 10. DAgger and interactive imitation learning | Notes · Slides · Practice |
| 6 | 11. Reinforcement learning for feedback control | Notes · Slides · Practice |
| 7 | 12. Control stacks, locomotion, and the course project | Notes · Slides · Practice |
Practices
- 1. A reproducible MuJoCo control loop
- 2. Differential IK and torque control
- 3. Cart-pole and quadrotor LQR
- 4. MPC and weighted stochastic shooting
- 5. Behavior cloning and DAgger
- 6. Tabular reinforcement learning
- 7. Reproducible control project
Assessment
Practices 20%, take-home 1 25%, take-home 2 25%, project 30%. See the assignment briefs and rubrics.
This is original teaching material. The source map records how each topic connects to the linked CMU and MIT courses and to MuJoCo APIs.