Transformer-based Multi-agent Reinforcement Learning for Multiple Unmanned Aerial Vehicle Coordination in Air Corridors

Animation on torus-cylinder-cylinder-torus

D3MOVE_v4.py for visualization of UAVs coordination in air corridors.

Air Corridor Modeling

UAVs need to traverse several air corridors to reach their destinations. Air corridors are modelled as cylinder and partial torus.

Cylinder and Torus

RL Training

Network Structure

• H(), embedding layer, normalizes the input values and standardize the input dimensions.
• G(), transformer layer, deals with stochastic neighbors information
• F(), actor-critic network combined.

Training File

main.py