Dynamic-TD3 Algorithm Enhances UAV Path Planning with Obstacle Prediction

A novel algorithm, named Dynamic-TD3, has been introduced for the path planning of unmanned aerial vehicles (UAVs) operating in intricate environments. This framework tackles the challenge of balancing safety and exploration in deep reinforcement learning by framing navigation as a Constrained Markov Decision Process (CMDP). It employs an Adaptive Trajectory Relational Evolution Mechanism (ATREM) to understand long-term objectives and utilizes a Physically Aware Gated Kalman Filter (PAG-KF) to reduce sensor noise. The policy, which adheres to dual criteria, ensures mission efficiency while adhering to strict safety constraints through Lagrangian relaxation. This research can be found on arXiv under the identifier 2605.00059.