Arb Chameleon: PPO Cross-DEX Arbitrage Agent
This model is a Reinforcement Learning agent trained to identify and execute atomic price spreads across decentralized exchanges (DEXs). It uses Proximal Policy Optimization (PPO) to make trading decisions in a simulated environment that accounts for gas volatility, slippage, and transaction fees.
Model Details
- Model Type: Proximal Policy Optimization (PPO)
- Library: Stable-Baselines3
- Environment: Custom Gymnasium environment (
ArbEnv) simulating multi-DEX arbitrage. - Goal: Maximize net profit while minimizing failed transactions and gas waste.
Intended Use
- Research & Education: Understanding how RL can be applied to DeFi arbitrage.
- Simulation: Testing trading strategies in a controlled, realistic environment.
- Inspiration: A starting point for building more complex, production-ready trading bots.
Limitations & Risks
- Market Dynamics: The model is trained on historical and simulated data. Real-world market conditions can change rapidly.
- Gas Costs: While the model considers gas, sudden spikes in network congestion can lead to unprofitable trades if not handled by the execution layer (e.g., Flashbots).
- Execution Risks: This model only provides the strategy logic. The actual execution layer (smart contracts) must be robust and secure.
Training Data
The agent was trained on a universe of assets including ETH, BTC, and various stablecoins across major L1 and L2 chains (Ethereum, Arbitrum, Base, Polygon).
How to Use
To load and use this model, you will need the Arb-Chameleon repository and the following dependencies:
pip install stable-baselines3 gymnasium numpy torch
from stable_baselines3 import PPO
# Note: You need the project's source code to define the environment
# from rl.src.env import ArbEnv
# Load the model weights
model = PPO.load("final_model.zip")
For more details and the full source code, visit the - **GitHub Repository**: [Arb-Chameleon](https://github.com/sdi1400258/Arb-Chameleon)
- Downloads last month
- 3