Dynamic-TD3: A Novel Algorithm for UAV Path Planning with Dynamic Obstacle Trajectory Prediction

A team of Chinese researchers (Wentao Chen, Jingtang Chen, Mingjian Fu, Tiantian Li, Youfeng Su, Wenxi Liu, Yuanlong Yu) has introduced Dynamic-TD3, a novel algorithm that tackles the long-standing safety-exploration dilemma in autonomous drone navigation. Traditional deep reinforcement learning (DRL) methods either use soft penalties that encourage risky trial-and-error or rely on constraint-based approaches that degrade under sensor noise and intent uncertainty. Dynamic-TD3 reframes navigation as a Constrained Markov Decision Process (CMDP) and incorporates two key innovations: an Adaptive Trajectory Relational Evolution Mechanism (ATREM) to capture long-range movement intentions of dynamic obstacles, and a Physically Aware Gated Kalman Filter (PAG-KF) that filters out non-stationary observation noise while preserving physical consistency. The resulting state representation drives a dual-criterion policy that balances mission efficiency against hard safety constraints via Lagrangian relaxation.

In experiments involving aggressive dynamic threats, Dynamic-TD3 demonstrated superior collision avoidance performance, reduced energy consumption, and smoother flight trajectories compared to baseline methods. While the paper (6 pages, 5 figures, submitted to arXiv Robotics) does not release code or real-world flight tests, the framework shows strong potential for real-time deployment in cluttered environments like urban deliveries, infrastructure inspection, and emergency response. By combining physical awareness with reinforcement learning, Dynamic-TD3 moves toward closing the gap between simulation and reliable real-world operation—a key requirement for scaling commercial drone services.