A collaborative inspection system composed of quadruped and flying robot for crack segmentation in tunnel environment

Abstract

Advanced robotic systems involving unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) have shown advantages in detecting damaged infrastructures, such as bridges, buildings, and nuclear power plants. Due to their ability to leverage varied vantage points, a robot team of heterogeneous UAVs and UGVs could collect full space information of the structure with wider coverages and higher robustness compared with that from each robotic platform alone. However, critical issues still exist when applying the UAV-UGV inspection system in the tunnel environment. One major problem is that the lack of light and GPS signals may lead to a higher probability of UAV collisions. Another problem is that the road surface in the tunnel is generally uneven with many obstacles and puddles, which could hinder the travel of traditional wheeled or tracked UGVs. To address the above issues, a novel collaborative robotic system was proposed that consists of a quadruped robot with strong obstacle-striding abilities and a collision-resilient UAV. In the proposed inspection framework, the quadruped robot follows a pre-planned path and scans the tunnel lining with the vision sensor. Based on the scanning results, the UAV equipped with a protective cage would be launched from the helipad of the quadruped robot to perform a closer inspection. To correct its flight trajectory and protect the UAV, a GPS-free self-localization algorithm is constructed based on the data from the onboard computer, camera, and inertial measurement unit (IMU). Finally, to ensure that the tiny defects can be detected efficiently, a multi-scale feature fusion segmentation network with the attention mechanism was applied to images taken by the UAV. The performance of the system is validated against a field test, which demonstrates the feasibility of developing and deploying a collaborative inspection system using quadruped and flying robots for tunnel inspection.