CoralPlan guides underwater robots for better coral inspection views
CoralPlan: Observation Skill Selection and Execution for Underwater Robotic Inspection
Robotics
Summary
Underwater robots need to capture the right views to inspect complex coral structures properly. The authors present CoralPlan, a system that chooses and carries out specific camera movements based on the task and current image. The system was tested with simulations and real robots in a pool, showing it can complete planned observations about 78% of the time. This work helps link how well robots pick their views to how well they perform in underwater tasks.
What this means in practice
- •For marine robotics teams: Enable underwater inspection robots to select and execute camera motions tailored to coral inspection tasks for improved data capture.
- •For environmental monitoring teams: Support monitoring of underwater ecosystems by ensuring robotic inspections maintain proper target tracking and coverage.
Authors
Yuer Gao, Yu Zhao, Yi Cai
Abstract
Underwater robotic inspection depends on acquiring views that reveal task-relevant structure. For a structurally complex coral colony, recognising the target is only the starting point: the robot must select and execute a viewing motion suited to the inspection task. We present CoralPlan, a vision-language system that selects an observation skill from a current camera image and task text supplied by an episode manifest. A shared motion interface executes orbit, patch, or survey as target-relative trajectories; the remaining plan fields provide operator guidance. Observation completion requires target keeping and primitive-specific coverage, while joint success also requires selection to match the recorded reference. We evaluate this interface in 144 simulated episodes and 36 matched simulation-hardware pairs. In a clear-water pool with external target-reference poses, hardware observation completion reaches 77.8% and joint success reaches 63.9%. The experiments identify both reference-mismatched completions and incomplete observations after a matching skill selection. These results connect observation-skill choice to measurable underwater execution outcomes and identify where task-directed acquisition succeeds or fails.