Motion Planning and Control for Underwater Vehicles Operating at Aquaculture Facilities

Abstract

Autonomous mobile robotic systems are at an increasing rate making the transition from laboratories and controlled environments to the complex, dynamic, and unstructured real-world. A crucial part of enabling this is the ability to perceive, plan and react to changes in the environment. Aquaculture is an example of a challenging and complex environment for autonomous underwater vehicles to operate in. This domain is characterized by harsh sea loads, flexible net structures that deform and move with ocean currents, and dense populations of fish. Current state-of-the-art practices rely on divers or human operators to pilot remotely operated vehicles for inspection and maintenance purposes, and there is currently a research effort in developing autonomous vehicles that can reduce costs, increase the operational weather window, and lower the risk of human errors and injuries. The objective of this thesis is to address this need for increased vehicle autonomy in aquaculture by developing and validating new methods that enable safe and controlled navigation in dynamic environments. The main contribution is a method that estimates the shape of a flexible aquaculture net pen using the vehicle’s onboard sensors. The net pen is modeled numerically, and its shape is updated with distance measurements from a forward-looking Doppler velocity log. This approach contrasts present methods that either assume a rigid structure, rely on instrumenting the net pen or knowledge of the current flow, or methods that only estimate the vehicle’s position relative to a local part of the net pen, and thus enable safe path following relative to the net pen. Another contribution is the development of a motion planner for dynamic environments. The method proposes several extensions to the elastic band method that enable safe navigation in the presence of fast-moving obstacles. By considering swept-out volumes of both the vehicle and obstacles, potential future collisions are predicted, and a functional is introduced that forces the planned trajectory away from these collisions. Further extensions are also proposed to give robustness to uncertainty in state and obstacle motion. Utilizing the aforementioned methods, a new method that enables autonomous navigation relative to net pens is proposed. First, waypoints are defined and updated relative to its estimated shape. Further, a functional is introduced to the motion planner that forces the planned trajectory to converge to a desired distance to the net pen, such that the vehicle can maintain a safe distance and inspect the net while reaching the desired waypoints. It has been a goal to validate the proposed methods in realistic conditions. To this end, a motion control system for remotely operated vehicles has been developed. Further, the elastic band motion planner, the net pen estimation method, and the net-relative navigation method have been validated experimentally at full-scale fish farms with promising results. The thesis also explores additional methods to increase the autonomy level of underwater vehicles. In particular, a new path following guidance law is proposed that takes inspiration of sliding mode control to guide a vehicle to a path with finite-time convergence. Furthermore, a new method for estimating the discrete mode of an amphibious robot is proposed. This robot has different locomotion styles for swimming in water and crawling on land, and thus is reliant on identifying the discrete transition between these modes, which is done using a salted Kalman filter.