GATSBI: An Online GTSP-Based Algorithm for Targeted Surface Bridge Inspection

TL;DR

Evaluation of the GATSBI algorithm reveals that targeting the inspection to only the segmented bridge voxels and planning carefully using a GTSP solver leads to a more efficient and thorough inspection than the baseline algorithm.

Abstract

We study the problem of visual surface inspection of a bridge for defects using an Unmanned Aerial Vehicle (UAV). We do not assume that the geometric model of the bridge is known beforehand. Our planner, termed GATSBI, plans a path in a receding horizon fashion to inspect all points on the surface of the bridge. The input to GATSBI consists of a 3D occupancy map created online with LiDAR scans. Occupied voxels corresponding to the bridge in this map are semantically segmented and used to create a bridge-only occupancy map. Inspecting a bridge voxel requires the UAV to take images from a desired viewing angle and distance. We then create a Generalized Traveling Salesperson Problem (GTSP) instance to cluster candidate viewpoints for inspecting the bridge voxels and use an off-the-shelf GTSP solver to find the optimal path for the given instance. As the algorithm sees more parts of the environment over time, it replans the path to inspect novel parts of the bridge while avoiding obstacles. We evaluate the performance of our algorithm through high-fidelity simulations conducted in AirSim and real-world experiments. We compare the performance of GATSBI with a classical exploration algorithm. Our evaluation reveals that targeting the inspection to only the segmented bridge voxels and planning carefully using a GTSP solver leads to a more efficient and thorough inspection than the baseline algorithm.

Figures (12)

• Figure 1: An example of $View$. The green box depicts the face of a bridge voxel, the blue cone depicts the viewing cone, and the red band on the cone depicts the viewing distance.
• Figure 2: Example of $View$ where bridge voxels can be inspected.
• Figure 3: Flow diagram of GATSBI. The algorithm creates an occupancy map of the environment using incoming LiDAR scans. Then, it segments the points corresponding to the bridge into another point cloud using the RGB camera images. It then makes another occupancy map of only the bridge using the segmented point cloud. GATSBI uses both the environment and bridge occupancy maps to generate viewpoints, points in free space where the UAV can inspect the bridge. It sends these to the GTSP instance to make a tour and then a local path planner to get the flight path.
• Figure 4: Full voxel map containing $V_{BI} \cup V_{BN} \cup V_O$.
• Figure 5: Example GTSP setup. Each inspectable bridge voxel can have multiple potential inspection viewpoints (vertices). All the vertices for a single bridge voxel are clustered together. The edges between these vertices are initially their Euclidean distance.