RI-RRT*: a riemannian metric-based path planning algorithm for autonomous robots in nuclear environments

Robotic missions in nuclear radiation environments require path planning that not only ensures obstacle avoidance and minimal path length but also reduces cumulative radiation dose to guarantee mission safety and system longevity. In nuclear facilities, the cost field is dominated by heterogeneous and nonlinear radiation distributions, posing unique challenges not addressed by conventional Euclidean-based planning algorithms. This paper proposes a novel Riemannian Informed Rapidly-exploring Random Tree Star (RI-RRT*) algorithm, which integrates Riemannian geometry into sampling-based motion planning. The environment is modeled as a Riemannian manifold, embedding both radiation intensity distribution and obstacle locations into a unified metric tensor, allowing path costs to naturally reflect environmental risks. Unlike conventional Informed RRT*, RI-RRT* employs Riemannian-informed ellipsoidal sampling, preventing the premature exclusion of optimal low-dose paths. Two improvements are introduced: ;(1) Radiation-aware two-stage sampling, which switches from goal-biased sampling for rapid initial solution generation to Riemannian-informed sampling for efficient convergence; and ;(2) Radiation-constrained redundant point elimination, which smooths paths, reduces dose, and enhances executability. Simulation results in Geant4-modeled radiation fields demonstrate that RI-RRT* reduces cumulative radiation dose by 3.25%–12.19% compared with Informed RRT*, GB-RRT*, PRM, and APF-PRM, while achieving shorter and smoother paths. Real-world UAV experiments confirm the method’s feasibility and robustness.