A risk-aware frontier planner with sparse 3D observations for autonomous inspection in new-energy facilities

Abstract

Inspection robots can support routine monitoring in photovoltaic fields, battery storage sites, and other new-energy facilities, but these scenes combine repeated equipment rows, narrow service aisles, and partial occlusions. This paper studies a risk-aware frontier planner for unknown-facility inspection from simulated sparse 3D observations. Instead of reconstructing a complete 3D model, sparse depth/LiDAR-like returns are projected to a 2D occupancy grid for local planning. Each frontier candidate is ranked using travel distance, local information gain, and a clearance-related obstacle risk term. The selected frontier endpoint is then reached by a risk-weighted A* planner that favors collision-free trajectories with larger separation from energy equipment. In four procedurally generated facility layouts, the method keeps mapping coverage comparable to the compact sampled Frontier-A* baseline, shortens the average trajectory, and improves both clearance and near-obstacle exposure relative to risk-only and information-gain frontier baselines. The result is a reproducible simulated prototype for intelligent sensing and autonomous inspection in complex new-energy environments.