Feedback-Linearizing Control for Velocity and Attitude Tracking of an ROV with Thruster Dynamics Containing Input Dead Zones Jordan Boehm 1 , Eric Berkenpas 2 , Charles Shepard 3 , and Derek A. Paley 4 Abstract— This paper presents a dynamics and control framework to accomplish six degree-of-freedom tracking of attitude, velocity, and rotational rate setpoints for a remotely operated vehicle with nonlinear thruster dynamics. The thruster dynamics contain input dead zones that complicate linear state feedback control design, and are compensated with nonlinear control strategies, specifically feedback linearization. Modeling the thruster dynamics in the control design mitigates the input dead zones. Simulations with experimentally obtained thrust parameters show improved reference setpoint tracking when compensating for the thruster dynamics. I. INTRODUCTION Remotely operated vehicles (ROVs) are widespread and versatile, being applicable to deep-sea exploration and min- ing [1], marine research [2], hull inspection [3], and wreck- age surveying [4]. To accomplish these tasks, ROV control is typically accomplished through a variety of methods ranging from direct human-in-the-loop control to autonomous, logic- driven control [5]. Controllers for autonomous or semi- autonomous operation have been designed through a vari- ety of feedback frameworks, including feedback lineariza- tion [6], [7], robust control [8], [9], and adaptive control [9], [10]. Most ROV operations are accomplished by semi- autonomous or full human control, whereby direct com- mands from an operator are either processed by a controller or fed directly to individual thrusters [5]. Direct-controlled ROVs typically have orthogonal thruster configurations that allow for intuitive translations from commands to thrusts, but such actuator placement can complicate the vehicle design. As a result, fewer thrusters are often used, thus limiting maneuverability of the ROV [5]. To maintain generality, we analyze an ROV that has a specific thruster placement configuration to accomplish fully actuated control. An auto- stabilizing control system is assumed to process user com- mands into setpoints. J. Boehm is supported by a graduate research fellowship from the National Geographic Society. 1 Jordan Boehm is a graduate student in the Department of Aerospace Engineering at the University of Maryland, College Park, MD, 20742, USA. jboehm77@umd.edu 2 Eric Berkenpas is the director of the Exploration Technology Labora- tory at the National Geographic Society, Washington, DC, 20036, USA. eberkenp@ngs.org 3 Charles Shepard is the lead mechanical system designer of the Explo- ration Technology Laboratory at the National Geographic Society, Wash- ington, DC, 20036, USA. mshepard@ngs.org 4 Derek A. Paley is the Willis H. Young Jr. Professor of Aerospace Engineering Education in the Department of Aerospace Engineering and the Institute for Systems Research, University of Maryland, College Park, MD, 20742, USA. dpaley@umd.edu This work is presented with relevance to the application of ROVs to aquatic imaging, the primary function of an ROV under development by the National Geographic Society (NGS) shown in Fig. 1. Underwater filmmaking requires smooth setpoint tracking with human-in-the-loop operations. Reference setpoint attitudes and velocities are typically generated through user input and, for complicated thruster configurations, controllers are capable of effectively tracking commanded trajectories. Often ROVs maintain only active closed-loop control of three or four degrees of freedom, while allowing roll and pitch parameters to be passively stabilized by relying on the natural stability of the vehicle due to the relative locations of the centers of gravity and buoyancy [5], [7], [8], [11]. However, for the purposes of deep-sea imaging, it is useful to have full user control of all attitude parameters, similar to a multi-rotor aerial drone, in order to obtain the desired cinematic effects. To enhance controller performance and reduce limit-cycle behavior, actuator dynamics are accounted for in the con- trol design [8], [12]. A variety of methods for modeling thrusters for underwater vehicles have been developed in previous work. A two-state axial flow dynamic model [13]– [15] accounts for thrust overshoot but is limited to uni- directional flow characterization. A two-state rotational flow model [16] has no more model accuracy than the axial flow model. Lastly, a multi-directional axial flow model [17] requires a large number of parameters to be identified with extensive system testing. This paper expands upon a single- state voltage-driven thruster model presented in [8]. We consider an analog voltage signal (throttle) as the control input for the thruster dynamics, which also exhibit a dead zone nonlinearity. A single-state dynamic thruster model is valid for low-speed movement [8], [14], [15]. In previous work, robust and adaptive control techniques have been used for dead zone compensation in the absence of well-identified model parameters [18], [19]. This paper utilizes feedback linearization to compensate for nonlinear- ities in thruster dynamics, because high-quality propeller speed, thrust, and torque data obtained from a six-axis Gough-Stewart platform load cell (Fig. 2) are available [20]. Other techniques [18], [19] for improving the robustness of feedback-linearizing methods are out of the scope of this paper. The contributions of this paper are (1) a nonlinear con- trol law for throttle-controlled thruster dynamics with in- put dead zones using experimentally obtained parameters; and (2) implementation of a feedback-linearizing and dead- zone-compensating thruster controller for the six degree-of-