An Introduction to Reinforcement Learning - Snake Your Way Out of a Paper Bag
summary
Introduces a hands‑on C++ implementation that builds a Q-learning table (std::map/std::vector keys) with tunable alpha/gamma and epsilon 10% exploration, uses SFML for SFML rendering, and trains on toy tasks: a 1D line reaches ~89/100 wins after ~20 runs, a 5×5 grid snake learns to collect apples and avoid a static obstacle, sometimes scoring up to mid‑30s; no neural nets are used (deep Q is mentioned but not implemented), Q‑table saving/loading is supported and the agent mostly works though it still makes random exploratory mistakes.
AI-written summary · about
chapters (11)
- 0:00Reinforcement learning intro
- 2:13What is RL and its history
- 6:04One-dimensional random walk
- 10:00Learning with lookup table
- 19:34Q-learning and Bellman equation
- 26:24From one to two dimensions
- 31:03Adding obstacles and dynamics
- 38:15Snake game transition
- 43:00Snake AI demo
- 49:17Summary
- 54:00Q&A
related talks
53mAn Introduction to Reinforcement Learning - Snake Your Way out of a Paper Bag
01h 01mReinforcement Learning: Pac-Man
49mML: Teaching machine to learn from its own mistakes (...)
01h 31mReinforcement Learning- Pac-Man
44mReinforcement learning: A gentle introduction
01h 31mWorkshop on developing Reinforcement Learning
33mBreakthroughs and the Future of (Deep) Reinforcement Learning
28mDeep Reinforcement Learning: theory, intuition, code