Autonomous underwater vehicles have a variety of applications such as environmental monitoring, search and rescue, ocean exploration, and fish tracking. One such class of these vehicles is gliding robotic fish, which realizes energy-efficient locomotion and high maneuverability by combining buoyancy-driven gliding and fin-actuated swimming. The goal of this dissertation is to endow gliding robotic fish with advanced control capability and autonomy, to facilitate their ultimate applications in aquatic environments.First, an overview of the gliding robotic fish platform GRACE is presented and design improvements for the third generation of GRACE are discussed. These include adding Iridium satellite-based communication for remote operation, making the robot more robust for ocean operation, and developing a miniaturized version (Mini-Glider) to enable rapid testing of functionality and control algorithms. Second, a backstepping-based trajectory tracking controller for the energy-efficient gliding-like motion of gliding robotic fish is proposed. The controller is designed to track the desired pitch angle and reference position in 3D space. In particular, under-actuation is addressed by exploiting the coupled dynamics and introducing a modified error term that combines pitch and horizontal position tracking errors. Two-time-scale analysis of singularly perturbed systems is used to establish the convergence of all tracking errors to a neighborhood around zero. The effectiveness of the proposed control scheme is demonstrated via simulation and experimental results. Next, incorporating observability into control schemes is discussed. Incorporating observability can enhance an observer's ability to recover accurate estimates of unmeasured states, minimize estimation error, and ultimately, allow the original control objective to be achieved. The use of control barrier functions (CBFs) is proposed to enforce observability and thereby encourage convergence of state estimates to the true state in output feedback control schemes. The proposed approach is compared to a model predictive control (MPC)-based alternative that optimizes a weighted combination of an observability surrogate function and the control objective. Motivated by the applications of fish tracking and navigating in GPS-denied environments, the problem of target tracking, when only the distance to the target is measured, is addressed. It is found that both approaches are comparable in terms of observability and estimation error, but the CBF-based approach has an edge in terms of computational efficiency. Experimental validation of the CBF-based scheme is conducted with a Mini-Glider. To complete this body of work, a strategy for the exploration of unknown scalar fields under localization uncertainty is proposed. The strategy hinges on the concept of the multi-fidelity Gaussian processes (GPs) and sampling-based motion planning for information gathering. It uses multi-fidelity GPs to approximate the environmental field by assigning location-measurement pairs to a particular fidelity based on the level of uncertainty in the location estimate. An informative trajectory planner is then designed that plans not only where the robot should go, but also what types of motion (e.g. swimming, gliding, etc.) the robot should use to best gather information for the reconstruction of the field. Experiments are carried out on a Mini-Glider for the task of mapping the light field in an indoor tank. The results show that using a multi-fidelity GP model provides a better reconstruction of the field in terms of the weighted mean squared error when compared to using standard GP regression, where the localization error is ignored.
Read
