It is said generally that the history of autonomous underwater vehicles (AUVs) started with SPURV which was developed by APL of the University of Washington at the request of US Navy in 1957. This was an untethered underwater vehicle with a total length of 3.1 m and a displacement of 430 kg driven by a mono-axis motor and a silver-zinc battery. It was able to run with a maximum depth of 3600 m, a speed of 2–2.5 kt and a maximum cruising endurance of 5.5 h. For control, 12 bit preset command was used to change the steps of azimuth and depth, and hardwired logic circuit was used in the controller. The azimuth was controlled by the offset from the initial azimuth at launch. The roll is secured by the static stability of the body. The temperature and conductivity sensors were installed on the nose section, and the data was recorded on a tape recorder. This vehicle was used for wide-area ocean observation by US Navy.1 A modern AUV, for example, the REMUS 600 developed by Woods Hole Oceanographic Institution having the same size scale as SPURV, has an inertial navigation system of ring laser gyro fully integrated with a Doppler velocity log (DVL) and acoustic positioning system. Autopilot software enables independent control of fins providing yaw, pitch and roll control, altitude, depth, and track-line following. Furthermore, it can be equipped with optional forward fins available to maintain a straight heading in a cross current. This alignment and stability is essential for optimizing the performance of a synthetic aperture sonar. Various kind of sensors like synthetic aperture sonar (SAS), side scan sonar (SSS), and multibeam echo sounder (MBES) can be equipped as the payload sensor. This can run for up to 24 h with lithium-ion rechargeable batteries and it is also possible to avoid obstacles by using a front sonar.2 In a half century, due to the progress of digital technologies, sensor technology, and battery technologies, AUVs have made significant progress and they have been used widely for oceanography, industries, and defense. One of the most representative examples of AUV's strengths in wide-area subsea search was for Air France Flight 447 which crashed off the coast of Brazil in 2009. Multiple REMUS 6000, which is equipped with SSS and cameras, were used in this search activity, and after about 2 months of operation, they finally discovered the wreckage of an aircraft lying about 4000 m deep seabed and contributed to the recovery of the flight and voice recorders.3 From the perspective of industry, the AUV industry has moved from prototype development for research and development to mass production of systems now. However, the number of AUV manufactures is still limited, for example, “Husin” of Kongsberg in Norway,4 “Remus series” of Hydroid in United States (now a subsidiary of Huntington Ingalls Industries), and “Bluefin series” of Bluefin Robotics in United States (a subsidiary of General Dynamics Mission Systems).5 Mitsubishi Heavy Industries, Ltd. (MHI) had developed R&D AUV such as Urashima6 for JAMSTEC, and currently is manufacturing OZZ -5 AUV production model for the Ministry of Defense in Japan.7 This article mainly describes the expected future operation of AUVs to motivate AUV research and development in the academic world. This section describes the typical operation of an AUV, which is a prerequisite for a concrete view of the control system. As its name implies, AUV is an autonomous underwater vehicle that does not have a tether cable for power supply and communication. After it is launched from the surface vessel, it basically navigates autonomously according to a predetermined route plan because its underwater communication capacity is very limited. The first AUV, SPURV, traveled through the “empty” ocean to measure sea water temperature and conductivity. Therefore, it did not have sensors to detect the surrounding terrain. On the other hand, the main mission of the modern AUV is to survey the seafloor structure, or objects such as unexploded ordnance (UXO) using an underwater acoustic sonar. Since the AUV navigates at a relatively low altitude, it navigates autonomously while recognizing the seafloor and obstacles using DVL. This can detect the vertical altitude, and the forward looking sonar, which can detect obstacles and the seafloor ahead. In addition, in order to cover the search area with on-board sensors, a typical mission pattern is to cruise at a certain depth or altitude in a certain pattern (typically a lawnmower pattern), and maintain a constant speed so that the sensor detection range covers the search area. This mission pattern is planned and preset as a route plan before launch. In order to acquire data of the sonar, especially in the case of the synthetic aperture sonar, it is necessary to suppress the attitude fluctuations and keep straight for a crossing tidal current using a forward rudder as well. The AUV basically runs along a preset route plan unless there is an obstacle or emergency. After the crushing of the mission pattern is completed, AUV returns to the planned position and is recovered by the surface vessel. The acquired data is transferred to the computer on the vessel, and maintenance work such as battery charging is conducted, and then, the operation is repeated. Since the amount of the underwater acoustic sonar data is enormous and the transmission by underwater communication is limited, the data is evaluated by the human for the first time after the data analysis on the vessel. The next route plan may then be corrected depending on the data acquisition results. It should be noted that launch and recovery, especially recovery, is troublesome and dangerous work at sea due to the lack of a tether cable for the AUV, and the limitations under rough sea conditions. This leads to mission standby on the vessel and a resulting cost. After the AUV undersea survey, ROV may be used for detailed observation and underwater work depending on the purpose of operation. This is because the resolution of acoustic sonar is not as fine as the visual method, although it is possible to detect an “object like the target.” So the ROV needs to identify and determine “the target” using a visual or optical method, which has finer resolution than acoustic. Conventional AUV operation such as seabed surveys, or unknown object is almost the same procedure as described. Unlike such conventional operations, innovative technological developments for more efficient operations have been carried out for some applications. One such development is AUV's underwater docking technology, which has already been tried in various research studies. This allows the AUV to dock with an underwater docking station, charging electrical power to the AUV and transmitting data from the AUV, continuously to operate without launch and recovery. This has the possibility of providing innovative efficient operation of the AUV. To realize underwater docking, there are technical problems of a proximity sensor, a docking device, and guidance and attitude control. In some docking systems, mono-axial propulsion AUV, which is difficult to cruse at low speed, is caught by pushing it into a docking cage. However, an AUV has protrusions such as Fin, GPS/Wi-Fi antennas, and acoustic communication devices and also has a rubber surface for acoustic sensors. So this approach is risky for the AUV itself. Therefore, in consideration of the future expansion of AUV operations to be described later, it would be more feasible to adopt an approach in which an AUV capable of low speed or hovering is guided to precise docking positions and attitude, stabilized, and safely captured. Furthermore, the docking technology also can be applied to automatic launch and recovery operations. In order to improve the search efficiency, it is easy to come up with multiple AUV operations. In fact, several AUV's were used in the aforementioned Air France 447 search. Although it is only necessary to allocate each search area, it is necessary to introduce the viewpoint of automatic optimizing route planning. This is in order to eliminate duplication and omission of search areas, to consider work procedures such as launch and recovery, to adapt and easily change the plan according to the search situation, and to improve the overall efficiency. Furthermore, it will be possible that the operation of multiple vehicles will reach the level of Swarm operation by the ultra-small AUV. Airborne drones have been demonstrated to control 1000 groups, however they are still nothing more than technical interesting in terms of practical operations. Innovative practical ideas are required. In addition to multiple AUV operation, it is necessary to automatically optimize by combining heterogeneous unmanned vehicles in consideration of the entire practical operation, such as by cooperating with an autonomous surface vehicle (ASV) and the ROV. Even in the conventional operation pattern, there are technical problems to improve the search efficiency of AUV, and some of them are introduced below. At present, the AUV only controls the operation of its mission sensors such as SSS according to the pre-determined route plan. However, it is also conceivable that the search will be more effective by changing the route plan autonomously in real time according to the state of data acquisition by the mission sensor. For example, it is conceivable that, as a result of the automatic detection of a target like object in real time by a wide-area search sensor such as SSS, the wide-area search can be temporarily suspended, and then circling around the target area to obtain multi-angle data. It is also conceivable closing to the target to obtain the detailed data by using a fine resolution sensor such as underwater camera. This type of advanced operation eliminates the need for a second time operation, such as the “reacquisition” operation by ROVs, after the operation of the AUV, and will greatly improve the overall operation. Of course, some operators want to avoid the AUV having unintended behavior, so it is necessary to consider that the operator makes the final decision by using underwater communications. Such advanced autonomous behavior requires not only real-time automatic recognition technology of the sensor technology, but also the AUV itself is capable of low-speed cruising or hovering for closing approach to the target object. Both local route planning for the closing approach and route replanning to restart wide area search are necessary in real time and optimally. Despite the innovative autonomy in the previous section, there is a need to cruise at a lower altitude at a lower speed that is difficult for control. This is in order to obtain more detailed acoustic data or to survey with a short-range high-resolution sensor such as an underwater camera or laser. The lower the altitude, the more meticulous the route planning, and the more precise tracking and attitude control, considering sensor conditions, are required for complex terrains. : Vehicle velocity (velocity and angular velocity for each three axes). : Mass and added mass property terms. : Coriolis and centripetal terms. : Hydrodynamic damping force terms. : Hydrostatic force terms related to the Euler angle : Generalized actuator force and moment terms. There have been problems with the AUV control, such as disturbances caused by tidal currents, uncertainty of the model characteristics (e.g., fluid dynamics coefficients) and sensor data, nonlinearity of the actuators. The control engineers have designed and implemented an acceptable control system against such problems by conventional PID methods, linearizing the equation of motion and separating the control plane: vertical (pitch and depth) and horizontal (yaw and roll). However, PID control will be no longer applicable, when the AUV mission requires various maneuvers over a wide range of speeds including hovering, and side thrusters have been added to the AUV. AUV control engineers are facing the problems, such as, onboard reroute-planning, precise nonlinear maneuvering control in a wide range of velocities, while satisfying optimality and constraints such as energy limitations, attitude limitations for acquiring sensor data and actuator limitations in addition to the conventional problems of disturbances and uncertainties in the model. As for the development process, model-based development has already become main stream of control, and it is not necessary to discuss the merits of model-based developments here. In addition to this, it is expected that the development environment, simulation environment, and implementation environment will be seamlessly linked for an efficient development environment. In addition, considering the verification in three dimensions, such as in the complicated terrain tracking, it is difficult to evaluate only by the conventional time history graph. Visualization tools such as three-dimensional animation are required. Furthermore, considering the direct sharing of development methods and results, and the indirect sharing through education and human resource exchange, standardization of development, evaluation, and implementation environments is strongly required. De facto standards have already emerged. In this way, the efficiency of the development process on desk work has been improved, however when it comes to field testing, it remains unchanged. Running a system for the first time in water without tether cables and relying solely on acoustic communication, is still a major risk. After the initial test is completed and the vehicle starts to run, it does not follow the simulation because the AUV's body characteristics, especially the hydrodynamic characteristics, have the uncertainties mentioned. After correcting the control gain by trial-and-error, it should finally run as intended. Previously, the evaluation items were less due to its simple mission. However, in future, when AUV missions will be complex and evaluation items will become larger, we will not able to do such trial-and-error testing by human engineers. It leads to a huge number of sea trials and costs can be enormous. To avoid these high cost situations, adaptive methods are desirable. Adaptive control methods have been researched in the past, and recently, AI or re-enforcement learning is a rapidly growing area. We expect it will contribute in minimizing the development cost and operational cost. In the field of aerial robotics, the spread of low cost GPS, MEMS gyros, and microcomputers, as well as the development of software and network technologies, has led to the remarkable success of quadcopters. Drone has becomes the synonym of quadcopter although it means all kinds of unmanned robot. Drone is rapidly expanding their application from industrial applications such as surveying, agriculture, and entertainment to individual hobbies. Land robots are also a huge industry in terms of autonomous driving. On the other hand, underwater robots require special know-how such as pressure resistant structures and watertight structures in an underwater environment, and they are difficult to be recover when water leaks and sinks due to a small mistake. In addition, much of the equipment on board is expensive, such as underwater acoustic devices. For this reason, only a few manufacturers and research institutes with many years of experience in underwater vehicles can produce them, and their use is still limited to some marine industries and to defense fields. However, when considering the maturity of the technology, it is important to expand the scope of the industry, and for that purpose, it is important to reduce the AUV cost itself and its operating costs. From the development process of AUV to the improvement of operational efficiency by the innovation and upgrading of the operation, and so forth, the role of the control system is huge. We conclude with the hope that research and development and education on AUV control systems will continue to develop. I would like to thank to Chief Editor, Prof. M. J. Grimble to give me a chance to contribute this article, and Prof. I. Yamamoto, to advise and review the contents. And I am grateful to my colleagues on discussion about AUV control for the future.