Multi-UAV autonomous collision avoidance based on PPO-GIC algorithm with CNN-LSTM fusion network.

Area of Science:

• Robotics
• Artificial Intelligence
• Aerospace Engineering

Background:

• Collision avoidance is critical for safe multi-unmanned aerial vehicle (multi-UAV) operations.
• Existing strategies often struggle with complex, dynamic environments and limited airspace.
• Deep reinforcement learning (DRL) offers a promising approach for autonomous control.

Purpose of the Study:

• To develop an autonomous effective collision avoidance strategy for multi-UAVs.
• To enhance the efficiency, robustness, and accuracy of collision avoidance algorithms.
• To leverage deep reinforcement learning for real-time decision-making in confined airspace.

Main Methods:

• Designed an end-to-end deep reinforcement learning (DRL) control strategy with a potential-based reward function.
• Constructed a CNN-LSTM (CL) fusion network for multi-UAV information feature interaction.
• Proposed the CLPPO-GIC algorithm by integrating CL with a generalized integral compensator (GIC) into an actor-critic structure.

Main Results:

• The proposed CLPPO-GIC algorithm demonstrated improved collision avoidance efficiency.
• The inclusion of LSTM networks and GIC significantly enhanced performance.
• The algorithm's robustness and accuracy were validated across diverse simulation environments.

Conclusions:

• The CNN-LSTM fusion network effectively integrates multi-UAV information for better situational awareness.
• The generalized integral compensator further optimizes the control strategy for collision avoidance.
• The developed DRL-based approach provides a reliable solution for autonomous multi-UAV collision avoidance.