Research on UAV Dynamic Trajectory Prediction Based on Deep Learning

Accurate trajectory prediction for Unmanned Aerial Vehicles (UAVs) is paramount for ensuring operational safety and efficiency in increasingly crowded airspace. However, conventional prediction models struggle to forecast the complex, non-linear paths generated by modern Reinforcement Learning (RL)-based navigation policies. To address this challenge, this paper introduces TC-Former, a novel hybrid deep learning architecture that synergistically integrates a Temporal Convolutional Network (TCN) with a Transformer encoder. The model is designed to capture both local temporal features and long-range global dependencies within UAV motion data. We developed a high-fidelity simulation environment using Gazebo and ArduPilot SITL to generate a comprehensive dataset of a UAV performing obstacle avoidance missions guided by a Twin Delayed Deep Deterministic Policy Gradient (TD3) agent. Experimental results demonstrate that TC-Former achieves state-of-the-art performance, attaining a Coefficient of Determination () of 0.9980 and an Average Displacement Error (ADE) of m. This significantly outperforms established baselines, including standalone TCN and Transformer models, validating that our hybrid approach provides a more robust and precise solution for predicting the trajectories of RL-driven autonomous systems.