The Swarm: Autonomous Boats Take on Navy Missions


(Photo by John L. Williams)

By Dr. Robert Brizzolara

What if the Navy could perform some of its toughest and most dangerous missions using a large number of small and inexpensive unmanned surface craft, instead of with a small number of large and very expensive manned platforms? With the cost of Navy ships going up and their numbers going down, the concept of teams of inexpensive unmanned surface vehicles (USVs) becomes not only interesting but increasingly relevant. It would invert the cost asymmetry presented by many threats today. In addition, a team of USVs could be more survivable (since the team could lose a few of its number and still retain its mission effectiveness), less detectable, and more effective for certain missions than individual manned vessels.

These USV teams are enabled by autonomous control, which means the craft are able to drive themselves under remote human supervision rather than operate with the human remotely driving the boat. Controlling large numbers of small boats with the latter method is often not feasible because of the limitations of communications range and operator situational awareness. Autonomous control greatly reduces the bandwidth required to operate the USV, and the amount of cognitive workload on humans. This will allow USVs to operate much farther away from the control station and allows human supervisors to control multiple USVs.

A distributed system of USVs is compelling because it can be developed using platforms the Navy already has by installing an inexpensive kit that converts these boats from manned to unmanned control. Small craft already are carried on Navy combatants and thus easily transported anywhere in the world in sufficient numbers to perform many useful missions. In addition, USVs could be used with other unmanned platforms to provide increased mission effectiveness. For example, USVs operating in conjunction with unmanned aerial vehicles would provide multidimensional situational awareness.


In August 2014, the Office of Naval Research, along with several partner agencies and commands, conducted an autonomous swarm demonstration that employed key technology enablers for USV swarms. The Naval Expeditionary Combatant Command was assigned overall tactical command for the demonstration. The Coast Guard provided traffic control boats and closed sections of the James River for three 30-minute periods per day from 11-14 August. The swarm demonstration technical team included: Spatial Integrated Systems, Inc. (overall execution of demonstration, behavior development, wireless network implementation, and implementation of autonomous control using the Control Architecture for Robotic Agent Command and Sensing, or CARACaS); Daniel Wagner Associates (implementation of the Decentralized and Autonomous Data Fusion System, or DADFS, and radar); Johns Hopkins Applied Physics Laboratory (DADFS); Pennsylvania State University Applied Physics Laboratory (radar processing software); Jet Propulsion Laboratory (developers of CARACaS and behavior development); Naval Surface Warfare Center Carderock Division (four USVs, boat preparation, and demonstration safety); and Naval Surface Warfare Center Dahlgren Division (one USV).

The USVs were employed in a straits transit scenario that included escort and attack phases. The scenario was conducted in the confined waterspace of the James River near Fort Eustis, Virginia, and is depicted in Figure 1 (see page 16). It included the five autonomous USVs and eight remote-controlled, high-speed maneuverable seaborne target boats, all unmanned. There was a high density of contacts that included the USVs, remote-control boats, a friendly force “highvalue” unit, an adversary force contact of interest (a Mark V special operations craft), and various traffic control and support craft. The boats escorted the high-value unit from the southern starting point in the James River through the channel to the north. The contact of interest was the surrogate for the opposition force coming from the east to oppose the high-value unit. The highlights of each phase are described in Figure 1.

Figure 1 Web

In Figure 1 above, the Swarm demonstration’s overall scenario is shown.

Of the five USVs that participated in the demonstration, two were 11-meter rigid-hull inflatable boats (RHIBs), one was an 11-meter small-unit riverine craft boat, one was a 7-meter RHIB, and one was a 7-meter harbor security boat. The use of different boat types illustrates the versatility of CARACaS. All of these boats are currently in Navy inventory, so existing boats can easily be converted into autonomous USVs.


The demonstration featured two key technical enablers for distributed systems of USVs that are being developed by ONR: the CARACaS autonomous control system, and the DADFS system for fusion of shared situational awareness data. Each USV was equipped with a CARACaS “stack” (a compact processing unit), a commercially available marine radar for perception, and a DADFS unit. CARACaS takes the situational awareness information provided by the radar and DADFS and plans a route to escort the high-value unit or to take an appropriate attack action, depending on the scenario phase, while avoiding obstacles. CARACaS has significant behavior-based autonomous control capabilities that were used in the swarm demonstration.

CARACaS has been under development for approximately 11 years. In 2004, ONR initiated a science and technology program to develop autonomous control for USVs performing complex missions in unpredictable and harsh environments. CARACaS advances well beyond the state-of-the art by being able to respond to dynamic situations and organic machine perception. It leverages past NASA investments in artificial intelligence for Mars Rover missions, and has already seen more than 3,500 nautical miles of onwater development, testing, and experimentation time. Functionally, CARACaS consists of two components: a perception engine and a behavior-based control framework that includes a route planner. Both of these were developed by the Jet Propulsion Laboratory. A key enhancement to CARACaS that enables multiple-USV operations is DADFS, which allows situational awareness sharing and fusion. DADFS is a combination of data fusion algorithms developed by Daniel Wagner Associates and the Johns Hopkins University Applied Physics Laboratory’s distributed blackboard system. In an ONR-sponsored project, the DADFS prototype for unmanned vehicles was developed to obtain contact/track data, create a common situational awareness on each vehicle node using collective sensor data, and synchronize the vehicles’ situational awareness.

The swarm event was structured to be both a demonstration of the USV autonomous control technology and a science and technology experiment. It demonstrated the use of five autonomous USVs to escort one vessel and attack another. The experimentation was focused on evaluating the performance of the autonomous control system. The system’s performance will determine the degree of trust that human operators will have in it—and ultimately its usefulness to warfighters.

One of the challenges associated with this event was that little existed in terms of procedures or processes to evaluate autonomous control systems for USVs, so we largely developed our own methodology for the demonstration. Key quantities were identified and measured, such as the frequency of human intervention the amount of communications bandwidth used by the control system.


There were two key aspects of this event that had a positive influence on the results. First, the confined waterspace and high contact density meant there were frequent interactions between the USVs and other craft or keep-out zones (such as shallow water, markers, or buoys) that required the USV to maneuver. This facilitated the collection of a much larger amount of data on the control system’s performance than otherwise possible. Second, there were no safety riders aboard the USVs. This was a departure from ONR’s usual model for developing autonomous control in which safety riders are on the USVs to take control in case of a malfunction. The lack of safety riders meant that the remote human operator was more likely to take control in a questionable situation. This helped reveal situations in which the remote human lacked necessary, why that intervention was necessary, and trust in the autonomous control system and will facilitate the development of approaches to increase trust.

We found that the predominant causes of the remote human operator taking control of a USV were related to maintaining sufficient buffers around the USVs. For example, the USVs occasionally violated these buffers because of perception issues (false detections that caused the unnecessary maneuvering) or route planning issues (insufficient precision relative to the close quarters in which the USVs were operating). We are pursuing further technology development to decrease both the frequency of human intervention in the operation of the USVs, and the demand for communications bandwidth.

Based on the results of the 2014 autonomous swarm demonstration, autonomous control of at least five USVs for escort and attack missions is feasible. Further technology development will result in a system that engenders increasing levels of trust from human operators and therefore has maximum usefulness to warfighters. There are numerous potential missions for teams of USVs, and those missions and the environments in which they must perform vary greatly in complexity. As trust in the autonomous control system increases, it will be used for more difficult and challenging tasks.

About the Author:
Dr. Brizzolara is a program officer with the Sea Platforms and Weapons division at the Office of Naval Research.