AtcEnv

Overview

Artificial intelligence has been declared successful in providing decision support in a variety of real-world applications. Many of these accomplishments have been made possible by recent advancements in reinforcement learning (RL) algorithms. In short, RL algorithms can be used to discover the best strategy (policy in the machine learning jargon) for a wide range of difficult tasks simply by learning from the experiences of the agent interacting with the environment. The policy is typically a neural network that takes as input the state of the environment as observed by the agent and determines the best action to maximize the return (cumulative discounted reward).

Recently, EUROCONTROL has implemented a reinforcement learning system for training Air Traffic Control (ATC) policies. The current system is composed of (1) a relatively simple ATC simulator that generates experiences, and (2) a learner based on the Proximal Policy Optimisation (PPO) algorithm that uses these experiences to continuously improve the policy.

In this project our aim is to further improve the ATC-ENV and take part in the conflict resolution for the EUROCONTROL Innovation Masterclass challenge.