NEAT-based 3D path planning for mobile robotic arms in NDT with offline inverse kinematics validation

Abstract

To address the challenges of coverage completeness and path executability in mobile robotic arm planning for industrial ultrasonic nondestructive testing (NDT), this study proposes a 3D path planning method that combines the Neuro-Evolution of Augmenting Topologies (NEAT) algorithm with offline inverse kinematics (IK) checking. First, a graph structure is built from the point cloud data of the target surface, and the NEAT algorithm is used to evolve an access strategy that simultaneously optimizes coverage, path smoothness, and path length. An offline IK validation step is introduced to pre-evaluate the reachability of each node using a standard solver in ROS. Based on the IK results, the node selection is further optimized to reduce the risk of execution failure. For nodes that are unreachable by the robotic arm, static position adjustments of the mobile chassis (Husky) are applied to help maintain overall path coverage. Simulation results on the ROS + Rviz platform across different surface geometries show that the proposed method achieves 100% surface coverage in all tested cases, with no collisions occurring during execution. It provides a practical solution for inspection scenarios where arm reachability and surface complexity present significant challenges.