A Survey of RRT* Variants and Enhancements for Robotic Path Planning

Sampling-based planners like Rapidly?exploring Random Trees Star (RRT*) have proven powerful for robotic path planning. However, practical deployments expose limitations in convergence speed, dynamic feasibility, collision?checking overhead, and adaptability to changing environments. This paper introduces a novel practice-oriented taxonomy and decision map that aligns RRT* variants with constraints (dynamic, kinodynamic, and safety) and computes budgets. First, we review guided sampling techniques, including Informed RRT*, bridge?test sampling, and region?biased strategies that focus computational effort on promising regions, accelerating convergence. Next, we examine kinodynamic extensions such as Kinodynamic RRT* and LQR?RRT*, which embed system dynamics and optimal?control heuristics directly into the tree growth, yielding smooth, dynamically?feasible trajectories for under actuated manipulators. We then explore collision?checking optimizations, from lazy evaluation to multi?resolution batching, which reduce expensive obstacle?testing calls without sacrificing optimality guarantees. Moreover, dynamic?environment variants (Dynamic RRT*, ERRT, and hybrid RRT*-D* Lite) are surveyed to demonstrate efficient incremental re-planning under moving obstacles. Finally, we discuss post?processing methods, including CHOMP and shortcut smoothing that further refine raw RRT* paths into execution?ready trajectories. By synthesizing these improvements, we identify open challenges and propose a unified framework integrating guided sampling, kinodynamic control, and real?time re-planning for next?generation robotic path planners.