Fusion-Guided Framework: Improving Sampling Performance via Potential Fields and Terrain Density

Path planning algorithms are a crucial component of robotic navigation. Among them, sampling-based algorithms such as RRT*, Informed-RRT* and BIT* are widely adopted in complex environments due to their property of asymptotic optimality. However, their reliance on random sampling results in low sampling efficiency in narrow or misleading obstacle scenarios, leading to longer planning times. To address these limitations, this paper proposes a novel Fusion-Guided sampling framework (FG), which flexibly integrates Artificial Potential Field (APF)-guided sampling for goal-oriented exploration, Probability Density Function (PDF)-based sampling for focusing on key geometric features, and uniform sampling for maintaining global exploration. Fusion-Guided sampling framework employs a probabilistic mechanism to dynamically switch between different sampling strategies, significantly improving both the initial path search and subsequent convergence. Experimental combination with RRT*, Informed-RRT* and BIT* demonstrate that the proposed method consistently achieves shorter initial path length and convergence time across three scenarios with varying characteristics.