Mission Planning and Safety Assessment for Pipeline Inspection Using Autonomous Underwater Vehicles: A Framework based on Behavior Trees

TL;DR

The paper tackles the challenge of flexible, safe autonomous underwater pipeline inspection by integrating Behavior Trees (BTs) with BehaVerify for offline safety verification, and ROS/DUNE execution paired with UNavSim-based simulation for online operation and testing. It introduces a DL-driven pipeline-detection algorithm using Side-Scan Sonar and YOLOX SubPipe to enable real-time localization and tracking, with BTs coordinating search, approach, and inspection phases and safety checks expressed as LTL properties verified in nuXmv. The main contributions are (1) a BT-based safety-assessment framework, (2) an onboard pipeline detection and inspection algorithm tightly coupled to the BT, and (3) a simulation-first validation setup in UNavSim/DUNE including a Leixões marina scenario for data recording and geospatial alignment. The results demonstrate end-to-end capability in simulation, showing safe mission progression under predefined safety properties, while acknowledging limitations in realism of SSS sensing and the need for broader safety coverage. Overall, the framework advances autonomous underwater mission planning by enabling formal safety verification, automated BT generation, and pre-deployment validation, reducing risk for real-world deployments; see $G((batteryLow=1) -> stationKeeping)$ and $G((status != m ext{_station} ext{ and } status != m ext{_surface}) -> RowsTask.active)$ as representative safety/progress checks.

Abstract

The recent advance in autonomous underwater robotics facilitates autonomous inspection tasks of offshore infrastructure. However, current inspection missions rely on predefined plans created offline, hampering the flexibility and autonomy of the inspection vehicle and the mission's success in case of unexpected events. In this work, we address these challenges by proposing a framework encompassing the modeling and verification of mission plans through Behavior Trees (BTs). This framework leverages the modularity of BTs to model onboard reactive behaviors, thus enabling autonomous plan executions, and uses BehaVerify to verify the mission's safety. Moreover, as a use case of this framework, we present a novel AI-enabled algorithm that aims for efficient, autonomous pipeline camera data collection. In a simulated environment, we demonstrate the framework's application to our proposed pipeline inspection algorithm. Our framework marks a significant step forward in the field of autonomous underwater robotics, promising to enhance the safety and success of underwater missions in practical, real-world applications. https://github.com/remaro-network/pipe_inspection_mission

Figures (7)

• Figure 1: The proposed framework for safety assessment and behavior tree design. The green charts indicate offline design, and the blue ones indicate online execution for the mission. BehaVerify automatically generates the formal model and the behavior tree. The behavior tree is then implemented within the ROS framework to execute the mission sent to DUNE (robot's mid-level control). It retrieves its current state, fed to the UNavSim unavsim simulator for moving the robot in the simulated environment accordingly. Meanwhile, the mission data is recorded through the UNavSim API.
• Figure 2: Pipeline Tracking. This scheme depicts two pipelines, represented in grey, which could also illustrate a single pipeline partially obscured by sand. The orange squares indicate the algorithm step. The diagram characterizes the three different maneuvers, each marked by arrows in distinct colors: the Rows maneuver in yellow, the GoTo maneuver in green, and the Tracking maneuver in blue. The green squares mark the pipeline's detected and stored latitude/longitude positions. Finally, the red squares represent the AUV's position at the last pipeline detection.
• Figure 3: Fragment of pipeline inspection's Behavior Tree. Cyan nodes are selector nodes; the left grey nodes on each selector subtree are check nodes, and the right nodes are action nodes. The orange node is a sequence node.
• Figure 4: Pipeline inspection's Behavior Tree. In addition to the functional requirements for a pipeline inspection mission, the BT includes safety checks and subscription tasks to ROS/IMC messages for later use in simulation. The different phases of our algorithm are organized under the priorities subtree; the failsafe_trigger sequence checks if a battery low warning or restricted zone warning is raised; in such a case, it executes an emergency surface maneuver, and stops the mission. The detect_pipeline sequence executes the search phase if the pipeline is not detected; otherwise, it executes a goto maneuver to approach the pipeline and start with the next phase. With the track_pipeline sequence, the AUV navigates, as long as the pipeline is detected, along the pipeline and stores the position where a pipeline segment is detected. Finally, with the inspect_pipeline sequence, we command the AUV to navigate all the identified pipeline segments for RGB data acquisition. The mission ends with the AUV going to the surface and executing a station-keep maneuver.
• Figure 5: The integrated simulation framework allows synchronization between DUNE and UNavSim. On the left is the Neptus interface, the primary control and monitoring tool for the DUNE system. On the right is the UnavSim simulator, displaying realistic renderings of the environment and the underwater robotic vehicle. The UnavSim interface displays the simulation of the RGB camera and the corresponding segmentation labels.