Robotic path planning for NDT using neuroevolution of augmenting topologies

Abstract

Path planning in ultrasonic non-destructive testing (NDT) must simultaneously optimise coverage integrity and motion efficiency under complex geometric constraints. However, traditional methods primarily focus on shortest-path computation, which struggles to adapt to complex topologies, particularly in inspection scenarios where surface continuity and movement costs vary. These methods often fail to balance path quality and obstacle avoidance. In this study, a neuroevolutionary framework is introduced that combines graph-based modelling with the dynamic topology evolution of the neuroevolution of augmenting topologies (NEAT) algorithm, aiming to improve adaptability in complex inspection environments. The method models the inspection environment as a weighted graph and employs neural networks to optimise the path generation strategy that ensures complete coverage while minimising redundant motion. The experimental results demonstrate that the proposed method achieves 100% coverage while improving path smoothness, reducing sharp turns and enhancing obstacle avoidance. Overall, it outperforms A*, Dijkstra and a multi-layer perceptron (MLP).