This repository has all the codes and sources of various RL algorithms that I have implemented.

A collection of deep and tabular reinforcement learning algorithms, including implementations of DQN, Dueling DQN, Q-learning, SARSA, n-step Tree Backup, and Monte Carlo.

Q-Learning Agent that Learns and Plays Alan Parr’s Traffic Lights Game

Benchmarking 11 tabular RL methods on a 6×6 FrozenLake grid under deterministic and slippery dynamics.

Matchbox-RL: A tangible reinforcement learning library for Python. Based on Donald Michie's 1961 MENACE algorithm using matchboxes and colored beads. Perfect for education and visualizing RL.

Reinforcement learning for reactor control under partial observability using Q-learning and SARSA(λ), with analysis of noise robustness and policy behavior.

OpenSpiel is a collection of environments and algorithms for research in general reinforcement learning and search/planning in games.

Implementation of Q-Learning, SARSA, and Dyna Q to allow an agent to navigate the FrozenLake-v0 environment from OpenAI.

RL was cheaper. The heuristic was safer. Neither was correct. POLARIS stress-tests operational policies under chaos, demand spikes, and black swan events asking one question: which policy survives when everything goes wrong? Built with constrained RL, Bayesian modeling, CVaR risk metrics, and a human-in-the-loop governor.

强化学习中文学习笔记与可复现实验仓库，涵盖 Q-Learning、SARSA、Monte Carlo Control 及 FrozenLake、CliffWalking、Blackjack 示例。