Toward next-generation autonomous structural health monitoring and inspection with humanoid robots

Humanoid robots have recently drawn increasing attention due to their anthropomorphic morphology, high degrees of freedom, and integrated multimodal sensing capabilities, and they have been explored in a range of robotics and engineering applications. However, their potential for structural health monitoring remains largely unexplored. Current indoor SHM practices still rely primarily on manual inspection or on mobile robotic platforms with limited adaptability and flexibility, which constrains inspection efficiency and operational coverage in complex indoor environments. Motivated by this gap, this study proposes RoboInspect, an end-to-end autonomous inspection framework that employs a full humanoid robot for indoor civil infrastructure inspection. RoboInspect comprises three key modules: (i) a reinforcement learning–based locomotion controller trained and deployed on humanoid hardware, (ii) an autonomous navigation and path-planning pipeline that leverages prior maps together with onboard perception, and (iii) a vision-based perception module for structural crack detection in indoor inspection scenarios. These modules are designed to operate in coordination to support repeated, goal-driven indoor inspection tasks. The framework is evaluated through two representative case studies conducted in a university building, covering first-floor common areas and a stair landing with an adjacent corridor. Under controlled experimental conditions, the system demonstrates reliable autonomous navigation along predefined inspection routes and effective crack detection, with a minimum detectable crack width exceeding 0.2 mm. These results provide preliminary evidence supporting the feasibility of humanoid robots as inspection agents for civil engineering infrastructure.