tabular-rl

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

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

A comprehensive research framework for Temporal Difference learning, implementing 9 state-of-the-art algorithms with theoretical analysis, empirical evaluation, and interactive visualization. Features custom environments, convergence proofs, bias-variance tradeoff analysis, and reproducible experimental pipelines for reinforcement learning research

Q-Learning agent that learns to navigate a 4×12 cliff-edge grid using ε-greedy exploration and Bellman updates — trained from scratch with NumPy + Gymnasium in 63 lines of Python.

Standalone Q-learning and SARSA implementations with FrozenLake ablation experiments.

Síntesis de multiplicadores binarios mediante RL tabular: agentes Monte Carlo, Q-learning y SARSA rellenan tablas de productos parciales verificadas en Verilog (Icarus, CPU) o evaluadas como tensores batched en GPU (PyTorch/CUDA).

RL was cheaper. The heuristic was safer. Neither was correct. POLARIS stress tests operational policies under chaos, demand spikes & 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.

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

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

Seeded tabular Q-learning for CartPole-v1 with corrected state discretization, deterministic evaluation, and fully recorded experiment artifacts.

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.