Mapping-Guided Task Discovery and Allocation for Robotic Inspection of Underwater Structures

TL;DR

The paper addresses autonomous underwater hull inspection without prior geometry by integrating SLAM-derived meshes with a decentralized task allocation (CBBA) framework. It introduces Map-TIDAL, which discovers inspection points (CIPs) from a mesh, prunes them based on proximity and visibility, and ranks them via keypoint scores within the camera field of view, enabling efficient, distributed task assignment among robots. In-water experiments demonstrate that Map-TIDAL achieves superior coverage with fewer uninspected regions and higher average views per voxel than Voronoi and boustrophedon baselines, validating its effectiveness under limited communication and challenging visibility. The work contributes a practical, SLAM-guided methodology for robust underwater inspection and outlines avenues for enhancement through multi-modal sensing and collision avoidance.

Abstract

Task generation for underwater multi-robot inspections without prior knowledge of existing geometry can be achieved and optimized through examination of simultaneous localization and mapping (SLAM) data. By considering hardware parameters and environmental conditions, a set of tasks is generated from SLAM meshes and optimized through expected keypoint scores and distance-based pruning. In-water tests are used to demonstrate the effectiveness of the algorithm and determine the appropriate parameters. These results are compared to simulated Voronoi partitions and boustrophedon patterns for inspection coverage on a model of the test environment. The key benefits of the presented task discovery method include adaptability to unexpected geometry and distributions that maintain coverage while focusing on areas more likely to present defects or damage.

Figures (7)

• Figure 1: Flowchart of the task discovery and task allocation processes.
• Figure 2: a. Inspection point (orange) generation from mesh (grey) using the centroid (blue) and normal (green). b. Comparison of inspection points (orange) to existing tasks (blue), point A has no tasks inside the radius of exclusion, point B has a task in the radius of inclusion, but the angle is outside of the angle of exclusion, point C has a task that is within both the radius and angle of exclusion, and will not be removed. c. Calculation of keypoint scores based on the camera field of view. The score is the total of the keypoints (green) that fall in the field of view for the inspection point (orange) normalized by the number of keypoints. d. Final pruning on remaining inspection points, done in alphabetical order tasks E and H are removed, the remainder become tasks.
• Figure 3: A. Left camera image with tracked keypoints overlaid in green and untracked keypoints in red. B. Kimera mesh with A. as texture. C. Initial tasks shown as arrows in red, mesh in blue, keypoints as rainbow dots. D. The 326 CIPs generated in teal. E. The 165 CIPs remaining after the first pruning in teal. F. The 49 tasks kept from their keypoint scores in green, with the remaining 116 CIPs in red. G. The 98 tasks kept after the final pruning.
• Figure 4: A. Top view of the test tank with dimensions. Solid lines show geometry used to generate inspection points for Voronoi and sweeping pattern methods, dashed lines show the edges of the voxelized region used to evaluate method effectiveness. B. Test tank with two modified BlueROV2s with the main robot at the center.
• Figure 5: Simulated coverage analysis for A Voronoi, B Map-TIDAL, and C sweeping patterns with task locations shown in blue and voxels colored more red with each time they are viewed.