Introduction
Aquaculture is an established industrial sector of progressively increasing importance for food production as an alternative that minimizes negative impact on climate change. Continuously growing demand of aquaculture infrastructure introduces important yet-to-be-addressed technological and logistical challenges, with reliance solely on human operators becoming unattainable in the future. Robotics and automation could become a leading force on expanding aquaculture operations and resolving scalability issues in a consistent and sustainable way [1] . However, exposed aquaculture settings are among the most challenging domains for robots to operate, due to uncertainty introduced from limited and low-quality sensor readings, lack of static reference caused by changing surroundings due the constant deformation of the fish cage structure, and necessity for real-time decision in environments with currents, surge, moving obstacles, and uncertainty. The ResiFarm project, aims to address such problems regarding scalability and safe operations in exposed aquaculture settings by providing fundamental technologies enabling underwater robots to operate safely, accurately, and resiliently. The potential of d ifferent types of autonomous robots is explored for the produced technology, such as ROVs and Underwater Swimming Manipulators (USMs), Figure 1.
Approach
Resilient autonomy depends on solving robustly two fundamental problems in robotics: State Estimation, which is the problem of localizing the robot with its surroundings, and Motion Planning, which is the problem of deciding on efficient and safe actions that the robot could take towards accomplishing a task. Unfortunately, both are becoming more challenging to solve in the underwater domain, and especially in aquaculture settings. The focus of this paper lies on safe and robust motion planning.
Even assuming perfect state estimation — a prerequisite for motion planning — safe autonomous aquaculture operations require addressing simultaneously and in real-time a combination of unique challenges, such uncertainty, unexpected surge and currents, potential control errors, moving obstacles, and moving deformable nets. Several motion planning approaches exist , such as sampling-based, lattice-based , and optimization-based motion planning . We focused on path optimization, due to its computational efficiency, quick replanning frequency, and guarantees, as shown in past work [2 ].
Such techniques can provide very quickly locally optimal solutions , which are sufficient given the limited on-board sensing range. Though, the most important advantage of these techniques is that user-defined cost functions could be implemented making them highly adjustable to different platforms, and for different tasks, such as for inspection [3].
Towards enabling autonomy, a novel framework , called ResiNav, was developed that guarantees safe navigation in the presence of motion errors due to currents, waves, or imperfect controls, localization and map uncertainty, and detected dynamic obstacles. ResiNav [4] deals with all these challenges in real-time in a holistic perspective by informing the path optimization process with the past experienced conditions, adjusting automatically the clearance needed. The necessary clearance is computed analytically with minimal information and provides a tight worst-case boundary that guarantees safety.
Results
Currently, several motion planning concepts have been tested i n simulation. The AquaVis [3] pipeline was applied for cage inspection . Virtual visual objectives were placed on the net , enabling the robot to inspect it from a desired proximity. The robot inspected the net , while also avoiding safely dynamic obstacles executing unknown trajectories in simulation, Figure 2. Additionally, ResiNav has been implemented and rigorously tested for different conditions, environments, and robot velocities in simulation . An example of one test case with strong side current is shown below in Figure 3. We speculate that, almost certainl y, a motion planner with such capabilities will be at the core of future safe and robust autonomy applied for ope rations in exposed aquaculture infrastructure.
Conclusion and Future Work
A collection of approaches is investigated to realize the ambitious goals of ResiFarm, towards robust motion planning, with a subset of them described briefly above. Future focus will be directed towards data collection and testing of robust motion planning concepts in real fish farms (infrastructure provided by SINTEF ACE) using ROVs and investigating extensions to different platforms, such as the Eelume USM. We aspire that the result from ResiFarm would provide fundamental technologies and insights to support scaling aquaculture operations in the future , safely, accurately, and resiliently.
Acknowledgements
This work was financed by the Research Council of Norway through the project
.
References
[4] M. Xanthidis, et al., 2023 , ResiPlan: Closing the Planning-Acting Loop for Safe Underwater Navigation, ICRA. (Accepted)