AI AGENTS ON NEAR
👁️
What is Agent Arcade?
Agent Arcade is an open platform where you can train AI agents to master classic arcade games using deep reinforcement learning. No prior machine learning experience needed - we handle the complexity while you focus on the fun part.
AVAILABLE GAMES
Pong
Simple paddle vs paddle
Beginner Friendly
~1 hour
Space Invaders
Defend Earth from aliens
Intermediate
~2 hours
River Raid
Control jet, manage fuel, destroy enemies
Advanced
~3 hours
Coming Soon
Mystery game
???
More games coming soon! _
QUICK START
agent-arcade ~ terminal
Prerequisites
- • Python 3.8 - 3.11 (3.12 has known issues, 3.13 not supported)
- • Linux, macOS, or WSL2 on Windows
- • 2GB free space, 4GB RAM recommended
- • Node.js & NEAR account (for staking)
$
# Clone the repository git clone https://github.com/jbarnes850/agent-arcade.git cd agent-arcade # Step 1: Create virtual environment chmod +x ./install.sh ./install.sh # Step 2: Activate virtual environment source drl-env/bin/activate # Step 3: Install dependencies chmod +x ./install_in_venv.sh ./install_in_venv.sh
✓ Virtual environment created ✓ Dependencies installed ✓ Atari ROMs installed ✓ Installation verified
$
# Train Pong agent with visualization agent-arcade train pong --render # Train Space Invaders with custom config agent-arcade train space-invaders --config models/space_invaders/config.yaml # Monitor training progress tensorboard --logdir ./tensorboard/DQN_[game]_[timestamp]
Training agent... Step: 10000, Mean Reward: -19.5 Step: 20000, Mean Reward: -17.2 ... Step: 1000000, Mean Reward: 15.8 Model saved to ./models/pong/[timestamp]/final_model.zip
$
# Login to NEAR wallet first agent-arcade wallet-cmd login # Evaluate trained model agent-arcade evaluate pong models/pong/[timestamp]/final_model.zip --episodes 50 # Evaluate with video recording agent-arcade evaluate space-invaders models/space_invaders/final_model.zip --render --record
Evaluating agent... Episode 1/50: Score 17 Episode 2/50: Score 19 ... Episode 50/50: Score 21 Mean Score: 19.4 Verification token generated: ~/.agent-arcade/verification_tokens/pong_[timestamp].token
$
# Submit verified score (requires evaluation token) agent-arcade stake submit pong 15 # View leaderboard agent-arcade leaderboard top pong # Check rewards and balance agent-arcade wallet-cmd status
Submitting score... Score verified ✓ Transaction: https://explorer.testnet.near.org/transactions/... Reward: 0.5 NEAR New balance: 10.5 NEAR
$
For detailed installation instructions and troubleshooting, check out the installation guide.
HIGH SCORES
DQN-MASTER21500
NEURAL-KING19800
DEEP-PLAYER17200
RL-WIZARD15900
Q-LEARNER14700