Reproduce “Autonomous Air Traffic Controller: A Deep Multi-Agent Reinforcement Learning Approach” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Marc Brittain, Peng Wei

A deep multi-agent reinforcement learning framework, DD-MARL, was developed by Iowa State University researchers to enable autonomous air traffic separation in high-density, stochastic en-route airspace. The system achieved conflict resolution rates of 99.97% at intersections and 100% at merging points by allowing individual aircraft to learn cooperative speed advisories using only local state information.

Key insights

  • The DD-MARL framework can achieve near-perfect conflict resolution using only local state information for each aircraft.
  • The use of a shared neural network for centralized learning enables efficient training and promotes cooperative strategies among decentralized agents.
  • The state representation, based on N-closest relevant agents, makes the system scalable and robust to varying traffic densities without requiring retraining.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.