Learning Controllers - From Engineering to AGI

This course focuses on teaching learners how to autonomously learn control with minimal prior knowledge using deep reinforcement learning methods. The course covers the challenges in real-world control scenarios, the collect & infer paradigm for Reinforcement Learning, and examples of agent designs that can learn complex tasks from scratch in simulation and reality. The skills taught include understanding the mission of DeepMind, the control problem in fusion energy, classical PID controller, classical reinforcement learning, optimizing inference, data collection methods, sensor exploration, and locomotion. The teaching method involves a lecture format with examples and discussions. This course is intended for individuals interested in reinforcement learning, neural networks, and learning control systems.