Instead of writing a rulebook for a robot, what if you could build a “brain” that learns on its own? This is the promise of Reinforcement Learning (RL), the branch of AI responsible for modern breakthroughs in robotics and gaming. In this Semilab, we will explore how machines learn from trial and error, just like a human mastering a new skill.
We will start with a simple "random walker" navigating a grid, using math to teach it how to find treasure. Over the week, we will upgrade this architecture from a basic cheat sheet (Q-Table) to a deep neural network (DQN). You will use Python and Gymnasium to train your own digital agents to navigate mazes, balance poles, and even land a lunar module.
Prerequisites: No prior AI knowledge is needed - just coding experience and a willingness to let your script fail repeatedly until it learns to succeed!