Problem
Research questions and friction points this paper is trying to address.
underwater robotic inspection
observation skill selection
coral colony
task-relevant structure
target-relative trajectories
CoralPlan: Observation Skill Selection and Execution for Underwater Robotic Inspection
🤖 AI Summary
This study addresses the challenge of autonomous observation viewpoint planning for underwater robots navigating complex coral structures by proposing CoralPlan, a vision-language system. The method innovatively incorporates vision-language models (VLMs) into observation skill selection, adaptively deciding among orbiting, patch, or survey strategies based on current imagery and task instructions. Furthermore, a unified motion interface is developed to seamlessly bridge high-level semantic decisions with low-level goal-relative trajectory execution. Extensive evaluations comprising 144 simulation trials and 36 hardware experiments demonstrate that, under clear-water conditions, the system achieves an observation completion rate of 77.8% and a joint success rate of 63.9%. These results effectively validate the feasibility of the proposed framework for task-oriented adaptive observation in challenging underwater environments.
📝 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.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Vision-Language System
Observation Skill Selection
Underwater Robotic Inspection
Shared Motion Interface
Sim-to-Real Evaluation
🔎 Similar Papers
No similar papers found.
