Part 7 · Learning from demonstration
Teaching a policy by showing it, and measuring honestly whether it learned.
Learn first
Section titled “Learn first”This part assumes the following. If any of it is new, learn it on AI Engineering from Scratch first:
- Statistics for Machine Learning
- Model Evaluation
- Multi-Layer Networks and Forward Pass
- Backpropagation from Scratch
- Loss Functions
- Introduction to PyTorch
- The Full Transformer — Encoder + Decoder
- Autoencoders & Variational Autoencoders (VAE)
- Diffusion Models — DDPM from Scratch
- Flow Matching & Rectified Flows
Lessons
Section titled “Lessons”| # | Lesson | Run type | Time | Status |
|---|---|---|---|---|
| 25 | Evaluating robot policies | CPU | ~40 min | ✅ done |
| 26 | Teleoperation, replay and data collection | CPU | ~45 min | ✅ done |
| 27 | Robot datasets | CPU | ~40 min | ✅ done |
| 28 | Behavior cloning | CPU | ~60 min | ✅ done |
| 29 | Distribution shift and DAgger | CPU | ~45 min | ✅ done |
| 30 | Multimodal actions: diffusion and flow policies | CPU | ~75 min | ✅ done |
| 31 | Action chunking and ACT | CPU | ~75 min | ✅ done |