Integrating LSTM-Based Target Prediction with TD3 for Enhanced UAV Pursuit-Evasion

To address the pursuit-evasion problem of unmanned aerial vehicles (UAVs), this paper proposes the TD3-TP algorithm framework. This framework integrates a target prediction network (TP Net) within the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, enabling UAVs to autonomously pursue dynamic targets in an obstacle environment. The TP Net, constructed with a long short-term memory (LSTM) neural network that has memory units and time series data processing capabilities, predicts the future position of the target based on its historical trajectory. This integration allows the pursuing UAV to generate its current action strategy based on both the current and future states of the evader. Comparative experiments and ablation studies demonstrate that TD3 combined with TP Net enhances the success rates and efficiency of UAVs in pursuit-evasion scenarios, more effectively fulfilling the requirements for UAV pursuit tasks.