Abstract
To address unclear multi-constraint coupling mechanisms, energy estimation errors, and poor dynamic feasibility in autonomous underwater vehicle path planning, this paper proposes an energy-efficient velocity-kinodynamic rapidly exploring random tree star algorithm. The method integrates a velocity–energy model, a velocity–curvature model, and a velocity–safe-distance model into a planning framework. Four targeted improvements are incorporated: a potential-field-guided sampling, an adaptive step-length expansion, a kinodynamic feasibility constraint, and a multi-objective cost function. Simulation experiments across four scenarios and five start–goal tasks show that the proposed method eliminates all curvature violations, reduces propulsion energy by up to 49.4% and improves mean minimum obstacle clearance by 34.2% and 47.4% over the two baselines. Ablation studies confirm that each module contributes to overall performance.