D3DWA: Adaptive Weight and Prediction-Horizon for Dynamic Window Approach via Dueling Double Deep Q-Network

This paper proposes D3DWA, an adaptive Dynamic Window Approach framework for robot local navigation. The method uses a Dueling Double Deep Q-Network to dynamically adjust the weight coefficients and prediction horizon of DWA during navigation. The key innovation addresses the limitation of fixed parameters in traditional DWA by enabling longer horizons for efficient motion in open areas and shorter horizons for feasible motion in narrow or cluttered regions. This research falls within the Multibody/Robotics domain of physical AI, focusing on robot motion planning in physical space.