Goal-oriented and active-detection based on rapidly-exploring random tree for path planning

Sampling-based path planners such as Rapidly-exploring Random Tree star (RRT*) have been widely applied to autonomous robot navigation, yet they suffer from unordered random tree growth and insufficient local environmental perception in dense-obstacle environments. To overcome these inherent drawbacks, this paper proposes a novel Goal-Oriented and Active-Detection Rapidly-exploring Random Tree* (GOAD-RRT*) planner. Globally, an adaptive probability-based node selection strategy weighted by estimated path cost is designed to dynamically adjust goal bias and suppress invalid random sampling, which alleviates the disordered expansion of random trees. Locally, a circular-neighborhood active detection module is integrated to adaptively explore feasible expansion directions around each node, enhancing local obstacle perception and generating smoother collision-free trajectories. The proposed planner inherits the native rewiring mechanism of RRT*, thereby preserving its asymptotic optimality. A set of ablation and comparative experiments conducted on multi-scale simulation environments verify that GOAD-RRT* achieves superior performance against other state-of-the-art sampling-based planners. Future research will focus on lightweight optimization of the detection module to extend the method to high-dimensional configuration spaces.