multi-agent-rl

This is a framework for the research on multi-agent reinforcement learning and the implementation of the experiments in the paper titled by ''Shapley Q-value: A Local Reward Approach to Solve Global Reward Games''.

Compact visual reinforcement-learning toybox in PyTorch, featuring small arcade-style environments for Q-learning, DQN, PPO, SAC, MCTS/self-play, and multi-agent RL

Multi‑agent RL for load balancing across heterogeneous servers, with a learned surrogate model to speed up training. Compares real‑simulator training vs. surrogate‑based training.

Dark Zero Point Genesis: PPO Latent World Models Under Thermodynamic Scarcity 256 agents. 128D Latent Manifolds. Zero supervision. Agents utilize PPO-clipped surrogate objectives. Survival = Predictive Error Coding (PEC) × Energy Efficiency across a 50/15 Seasonal Cycle.

A deep reinforcement learning system for optimizing bridge maintenance decisions across municipal infrastructure fleets, implementing cross-subsidy budget sharing and cooperative multi-agent learning.

Multi-agent Deep RL (Double+Dueling DQN + PER) for adaptive traffic signal control on a 2x2 grid, with SUMO and built-in simulators, emergency-vehicle preemption and a real-time dashboard.

Deterministic hex-grid soccer environment with two adversarial agents. Implements Q-Learning, Minimax-Q (via LP), and Belief-Q with online belief updates; trains in SE2G/SE6G to reduce state space and evaluates behaviors in the full environment with comprehensive visualizations.

A configurable 3D jump-chess framework with A/B geometries, multi-player rule switches, and RL/self-play training support.

Adversarial Co-Evolution of RL and LLM Agents: A framework for training high-performance PPO agents against Large Language Models in Gin Rummy, utilizing curriculum learning and knowledge distillation.

Hackable text MMO engine (WebSocket JSON) + ML research stack: DQN with curiosity, PBT, reward modes, curriculum, scripted baselines, eval harness, 50-agent conductor, live dashboard

An engineering-focused multi-agent reinforcement learning system for Texas Hold’em using PettingZoo AEC and a custom PyTorch PPO self-play setup.

A reproducible benchmark that races classical and learned mobile-autonomy stacks on identical tasks — navigation to robot soccer.

Project 3 of Udacity's Deep Reinforcement Learning Nanodegree Program

Hexapawn Game Engine Proper 3×3 board with pawn movement Strategic RL Agents Minimax with Alpha-Beta Pruning (depth configurable 1-7) Q-Learning with temporal difference updates Experience replay for efficient learning Epsilon-greedy exploration with decay Multi-level decision hierarchy (immediate threats → strategic planning)

🎯 Key Features 1. Flexible Game Configuration Adjustable grid size (3×3 up to 10×10) Customizable win condition (e.g., 5-in-a-row on a 7×7 board) 2. Two Competing RL Agents Agent 1 (Blue X) vs Agent 2 (Red O) Each has independent Q-Learning parameters Watch them evolve different strategies over time

MAPPO vs IPPO in adversarial multi-agent RL — alternating, population, and league self-play in a custom predator-prey gridworld

Multi-agent reinforcement learning (PPO) for adaptive traffic signal control, built with SUMO and Stable-Baselines3

크레이지아케이드 모작 + Reinforcement Learning ( DQN, PPO )

A specialized Reinforcement Learning (RL) project focused on multi-task mastery across 10 distinct gaming environments. General-Gamer-AI-Lite implements a lightweight multi-task agent designed to learn shared representations and transfer knowledge between varied game mechanics, from classic arcade challenges to strategic grid worlds.

Multi-Equipment CBM (Condition-Based Maintenance) optimization using Deep Q-Learning with cost leveling and scenario comparison. Advanced RL system with QR-DQN, N-step learning, and parallel environments for HVAC equipment predictive maintenance.