Multivehicle Autonomy Is Taking Off

Steinberg
By Dr. Marc Steinberg, Office of Naval Research

A single naval operator manages a diverse team of air and sea systems to find and track a suspicious vessel covertly. Rather than controlling individual systems, the operator directs the actions and behaviors of the team by specifying high-level mission objectives, constraints, allowable risks, and the types of decisions the system can make even on its own. Elsewhere, unmanned surface vehicles (USVs) escort high-value assets in a contested environment. One USV has unexpected engine problems, and the system automatically readjusts the plan to complete the mission. On a single 11-meter, rigid-hulled inflatable boat, a small explosive ordnance disposal team controls multiple undersea and surface vehicles. The users task the system through a common interface to conduct cooperative area searches, reacquisition, and identification tasks more quickly. Under the sea, networked, mobile autonomous systems sense the environment to support oceanographic research and tactical operations in situations where environmental conditions are highly variable. The systems work together on tasks ranging from following gradients to making decisions on how best to sample the environment based on high-level goals. On the shore, a user requests tactical intelligence services from a distributed group of autonomous systems with a simple app-like interface. Mission tasks are allocated among the group to search, monitor, and patrol in a poor communications environment. To do this, the user does not require detailed understanding of how the autonomy operates.

These may sound like futuristic concepts, but they actually represent real multivehicle demonstrations that leveraged technologies developed by the Office of Naval Research. These kinds of experiments have become increasingly common both within and outside the defense research community. Examples can be seen in university videos posted on YouTube and in even some commercial applications, such as warehouse robotics. Some naval experiments were done as part of fleet exercises such as Trident Warrior or Rim of the Pacific. Others were done with commercial off-the-shelf systems as part of standalone experiments or in more informal events such as the Naval Postgraduate School/Special Operations Command field experimentation program.

Beyond these examples, laboratory experiments are increasingly using inexpensive and expendable systems managed at a mission rather than individual vehicle or sensor level. The ultimate goal is robust self-organization, adaptation, and collaboration among highly heterogeneous platforms and sensors in a contested and dynamic battlespace. This kind of collaborative autonomy may be applied to missions as diverse as surveillance, electronic warfare, or logistics. Another goal is reducing the reliance on centralized command and control by using flexible decentralized systems that have the ability to get the right service to the right user at the right time. Collaborating systems may provide a survivable mission capability that cannot be easily defeated simply by losing individual platforms or sensors. They also may be able to do things that would be unaffordable or impractical otherwise. For example, there is a need for large, persistent, and pervasive sensing networks for battlespace awareness. However, it is likely that there never will be a sufficient number of people to manage and control so many platforms and sensors individually or to analyze and assess directly the large amounts of data such systems can collect.

The ultimate goal is robust self-organization, adaptation, and collaboration among highly heterogeneous platforms and sensors in a contested and dynamic battlespace.

Another important aspect of multivehicle autonomy capabilities is the extent to which the barriers to entry have become very low. Open architectures, such as the Robotic Operations System and the more maritime-oriented Mission Oriented Operating Suite, make many of the underlying component technologies easily available to anyone with the technical sophistication to leverage them. Online forums provide help for less-skilled individuals and groups, including open-source projects such as the DIYdrones.com autopilot and online designs that others can rapidly build, improve on, and share. Inexpensive commercial hardware, such as smart phones and video game motion sensors, combined with commercial vehicles provide the opportunity to scale up multivehicle systems in ways that not that long ago would have taken tens or hundreds of thousands of dollars to accomplish.

Desktop and other rapid manufacturing technologies such as 3-D printers, laser cutters, and desktop milling machines make it easy to add custom parts to the mix. Taken together, this puts the ability to create multivehicle systems into a variety of hands from states to small networked groups. This will only get easier as the relevant technologies advance.

The development of this kind of autonomous capability is expected to be an evolutionary process with systems increasingly able to complete missions independent of humans and cooperate more effectively with warfighters and other systems. Multivehicle collaborative systems build on decades of research advances in enabling technologies in areas such as autonomous planning and optimization, decision making, perception, navigation, and control. For example, ONR has conducted several on-water demonstrations of USV autonomous hazard avoidance, collision regulations compliance, and tactically relevant autonomous behaviors at ship speeds of up to approximately 30 knots. With this level of enabling autonomy, it becomes possible to consider new multivehicle/human interaction concepts that can leverage human strengths in cognitive skills, judgment, and tactical understanding in managing teams of systems, while freeing warfighters from tasks that can be achieved effectively by machines. Warfighters should be able to leverage their knowledge of the battlespace without needing to be experts on all the low-level workings of autonomous systems.

A number of programs have demonstrated supervisory control of fewer than 20 heterogeneous unmanned systems on the basis of high-level mission criteria such as objectives, priorities, risks, and constraints. What is feasible today can be highly dependent on the challenges of the environment, the complexity and time-criticality of the mission, and the level of skill and experience of the users. The operational tempo and environmental challenges of mine countermeasures, for example, has been a good match with the current state of multivehicle technology. Three of the major enabling technology limiters are the ability of individual vehicles to get on station, navigate through their environments safely in areas with other vehicles/units, and perform useful mission functions on their own with minimal human intervention. These exist to varying degrees in different mission domains, but there are technological limitations that depend on the difficulty of the environment, the complexity of the mission, the type of user, and the limitations of individual platforms. Further, the more mature approaches that have been demonstrated frequently depend on either centralized control or some degree of periodic centralized coordination or calculation.

One flight-demonstrated example of this is work on market-based approaches, where individual platforms or sensors “bid” on available mission tasks based on the “costs” (such as time or risk) involved in doing that particular task. In this case, much of the calculation is done at the individual system level. It does, however, ultimately require some centralized nodes to coordinate the task allocation. Another example demonstrated at sea was gradient following with collaborating autonomous undersea vehicles. The vehicles made their own decisions in a distributed way, but there was a need to share data to calculate the gradient at a centralized node and share that amongst the platforms.

Further in the future are research projects looking at highly decentralized concepts that can operate in very limited communications environments and on missions, such as force protection, where individual systems may need to act and adapt rapidly based on locally sensed information. For example, there are new approaches to the abovementioned gradient-following problem that require no explicit communication—only that the vehicles keep track of the positions of their nearest neighbors. One promising area ONR has been exploring is looking for inspiration from collective animal behaviors in nature.

The natural world provides many successful examples of organizational structures that are scalable and able to operate with limited communications, noisy sensing, and a lack of centralized control. In addition, many of these biological systems have evolved to adapt to considerable amounts of uncertainty in their environments. Inspiration may range from social insects to fish schools to smaller groups of more intelligent animals, such as wolves, dolphins, and nonhuman primates. The resulting engineered systems may be very scalable. Simple individual units with minimal communications can lead to versatile group behavior with minimal or no direct communication. It may be enough just to keep track of a few nearest neighbors visually or coordinate indirectly through effects on environment. In this kind of structure, collective decisions can arise from interactions of simple elements, and actions can be made based on local information. The system does not require a centralized common operating picture or direct sharing of mission data, as there may be mechanisms to make decisions at a collective level that do not require any individual to have all the data. These systems are not necessarily leaderless, but leadership can be an emergent and transient property that is not vulnerable to the loss of a command-and-control hierarchy.

The natural world provides many successful examples of organizational structures that are scalable and able to operate with limited communications, noisy sensing, and a lack of centralized control.

ONR is funding basic research programs that may someday support the ability of a hybrid human/robotic team to operate more effectively than a fully human or machine one by taking advantage of the things each does best. Moving from the current state of the art to more effective human/robotic teaming relationships may require reaching across currently separated disciplines to develop common frameworks that incorporate models of human and animal intelligence, new empirical and experimental techniques, and mathematically rigorous methods.

It also will be important to explore a broad range of human machine relationships. Current interfaces largely allow the human to task the robot, determine when the robot needs help, provide that help, and prevent the robot from doing something undesirable. If there is initiative on the robot’s side, it is often to request human help and not the reverse. To be effective team partners, robots will need to be able to recognize when the human needs help or information and then act in an appropriate, trustworthy, and nonintrusive way. Related to this is the general need for autonomous systems to have substantially improved teaming skills. These are critical for high-functioning human teams but may require solving some of the hardest problems in computational intelligence and robotics to achieve human capabilities in an engineered system. Luckily, not all robotics teaming problems require human levels of teamwork. Hybrid human/machine teams may enable new teaming arrangements that are not practical with all human ones. Thus, it is important to consider a broad range of capabilities rather than just focus on an ultimate goal of trying to duplicate the capabilities of highfunctioning human teams.

From today’s more mature approaches to a possible future of hybrid human/machine teams, there are a rich variety of autonomous systems and concepts that could be leveraged to deal with future naval missions. For some time to come, key limiting factors of autonomous systems are likely to be their capabilities for understanding relevant features of their environment, interpreting and fusing sensor data, navigating through uncertain and dynamic environments, dealing with limited and unreliable communications, and transforming complex mission requirements into both group and individual behaviors. Integrated man/machine systems also must be attuned to these limitations and take advantage of the different strengths and weaknesses of both humans and machines. The technologies being developed and demonstrated under a wide variety of ONR and naval research programs provide stepping stones toward control of much larger numbers of systems on more complex missions. We are still in the early stages in terms of understanding how best to use these technologies to transform future operations. With each new experiment, fleet exercise, and fielding of this type of technology, their future may become more clear.

About the author:

Dr. Steinberg is a program officer at the Office of Naval Research. He manages multidisciplinary autonomy research that cuts across different technical areas and mission domains, as well as applied research that focuses on autonomous air systems and heterogeneous multivehicle collaborative systems.

About Future Force Staff