rl_lib

A minimal reinforcement learning library written in Rust from scratch.

Features

• Core abstractions: Environment, Agent, Policy
• Spaces: Discrete, Box (continuous)
• Algorithms:
  • Q-Learning (tabular)
  • SARSA (tabular)
  • REINFORCE (policy gradient with MLP)
  • DQN (experience replay + target network)
• Included environments:
  • GridWorld
  • CartPole (simple physics)
• Minimal hand-written neural net backprop — no heavy DL frameworks required.

Quick Start

cd rl-lib
cargo test
cargo run --example q_learning_gridworld
cargo run --example reinforce_cartpole

Project Structure

• SPEC.md — interface and behavior definitions
• ARCHITECTURE.md — module layout and design decisions
• TODO.md — implementation checklist