Safe Reinforcement Learning Based Fixed-Wing UAV Trajectory Optimization and Collision Avoidance

This paper proposes a safe reinforcement learning (RL) method for fixed-wing unmanned aerial vehicles (UAVs) to reach their destination along an optimal collision-free path. The proposed method allows UAVs to avoid surrounding obstacles by adjusting their attitude based solely on locally sensed information. A key feature is a safe RL mechanism that manages collision risk by maintaining the received signal strength indicator (RSSI) between agents below predefined thresholds. Simulation-based performance comparisons with benchmark method demonstrate that the proposed method effectively prevents inter-agent collisions while maintaining a high success rate.