LEVEL 1 - 82 OF 115 STORIES Copyright 1985 EW Communications, Inc. Defense Electronics October, 1985 SECTION: ARTIFICIAL INTELLIGENCE; AI & ROBOTICS; Pg. 77 LENGTH: 2991 words HEADLINE: ROBOTS ON THE BATTLEFIELD BYLINE: By Ejiner; J. Fulsang III; Mr. Fulsang is the manager of program development for combat/physical security systems at Odetics Inc., Anaheim, Ca. HIGHLIGHT: Technological advances in imaging, processing and guidance may soon permit unmanned combat vehicles to travel from concept to reality. BODY: Artificial intelligence and robotics have captured world attention in recent years, and with good reason: when combined, they can yield a product that displays the apparent ability to act on its own accord. While this may initially bring to mind a rather comic image of little robots running around the house doing chores, a far more dramatic demonstration of the true capabilities of artificial intelligence and robotics can be seen in a useful tool of war: an autonomous battlefield robot that can receive and execute combat orders. Battlefield robotics are both operationally and technically feasible in the near future, but only if the programmatic hurdles can be overcome. Why Battlefield Robots? Four primary factors impel the development of autonomous battlefield robots. These include demographic and economic trends, the balance between quantity and quality in conventional forces, the increasing lethality of modern weapons, and the spread of international terrorism. U.S. Census Bureau statistics show that by the year 2000, the segment of the population between the ages of 18 and 25 -- the group that supplies the majority of military recruits -- will have decreased by 20 percent. Adding to the effect of this decrease is the fact that when economic activity picks up in the civilian sector, interest in military service declines sharply. The U.S. Labor Department forecasted an additional 2 million entry-level jobs by 1990, which means that the military can anticipate manpower problems as more eligible recruits are wooed by an increasing number of civilian jobs. Quantitative and qualitative trends in conventional forces are also a matter for serious concern. The traditional answer to the numerical disparity between NATO and Warsaw Pact forces is to counter the Soviets' quantitative advantage with a qualitative technical edge; however, it takes a great deal of quality to overcome a strong numerical advantage, and the Soviets have proved quite effective in adapting our technical advances for use by their own forces. What is needed is a new idea that, because of its lack of manpower content, is inexpensive enough to deploy in large numbers and is based on technology the Soviet Union has shown to be least capable of duplicating -- that of sophisticated computers. Recent combat experiences in the Middle East have shown the lethality of conventional weapons has increased significantly in recent years. Manpower-intensive military forces will always be vulnerable to a surprise attack, particularly one that involves the early use of nuclear, biological or chemical weapons. It is therefore essential that we develop the means to meet that initial onslaught with a resource we can afford to lose. There is also a significant peacetime mission for autonomous combat robots. The increasing frequency of terrorism and the growing sophistication of terrorist forces are growing concerns for the physical security community. Recent studies sponsored by the Defense Nuclear Agency (DNA) and others have concluded that robotics would be an excellent way to reduce the manpower-intensiveness and increase the efficiency of physical security operations. Based on these considerations, the Army's Armor and Infantry schools have submitted draft planning documents calling for autonomous battlefield robots with an anti-armor mission. The Marine Corps is attempting to develop the basic technology for an autonomous battlefield robot to carry out reconnaissance and surveillance missions. The DNA program is the most advanced, having followed up on its initial feasibility study with a system concept development request for proposals that should be awarded this year. In addition, the Defense Advanced Research Projects Agency (DARPA) is sponsoring an autonomous land vehicle project under the management of the Army's Engineer Topographic Laboratories (see sidebar). A Battlefield Robot Concept The size and configuration of a battlefield robot would be determined by its primary mission. With different weapons, sensors and platform configurations, it could be used for anti- armor, anti-personnel, anti-aircraft, indirect fire support or physical security missions. The example used here to illustrate the concept is an anti-armor robot. To simplify discussion of so complex a system, it is best to break it down into four major subsystems: the mission module, the platform, the guidance and control group, and the man-machine interface. The mission module is mounted on a rotatable parallelogram device, which simplifies reloading from the ammunition bustle in the rear and also permits the platform to remain in hull defilade as much as possible. The weapon system in this case is an anti-tank missile, but could also be a direct-fire cannon. Today's missiles can only deliver shaped-charge warheads, which are not very effective against some of the newer types of armor; these will eventually be replaced by hypervelocity missiles that can deliver kinetic energy warheads. For the present, however, kinetic energy warheads can be delivered only by direct-fire cannons. In order to be considered a credible weapon against all forms of threat armor, a direct-fire cannon must be at least 105mm. At that size, the recoil force requires an overall vehicle mass of at least 15 or 20 metric tons to avoid toppling the vehicle. A mission module based on a recoilless missile could be mounted on a much lighter vehicle -- perhaps as light as 2 or 3 metric tons -- and still have a basic load of 15 or 20 rounds. Of course, this presumes that a suitable fire-and-forget anti-tank missile will be available. The threat sensor, represented by the rectangular figure on top of the mission module, is mounted so that it rotates independently of the weapon bore. This permits the robot to maintain "module defilade" until the actual moment of firing, thus maximizing concealment. The type of sensor used for threat detection, classification and engagement must be able to perform those tasks without human assistance. This is not to say that human intervention will not be required; in fact, it will probably be many years before a robot will be permitted to seek out and destroy an enemy with complete autonomy. This will be even more true in cases where robots are closely intermingled with friendly forces on the battlefield. Some future battles will almost certainly be fought in adverse weather and at night; therefore, the threat sensor must be able to accomplish its missions under these conditions. This requirement brings up the issues of cost and complexity. Sensors that operate in the visual and near-infrared wavelengths tend to be relatively inexpensive and simple, but they also tend to be ineffective in low light or in the presence of atmospheric obscurants. On the other hand, sensors based on mid- and far-infrared perform a little better under poor atmospheric conditions, but usually require cryogenic cooling to enhance their signal-to-noise ratios. This cooling requirement greatly increases system complexity and cost. Although still considered to be in the developmental stages, millimeter-wave r adar promises to be a true all-weather sensor, with the added benefit of built-in range determination. However, millimeter-wave radar may prove too costly for deployment on each robot; this suggests that some sensors should be mounted on other platforms such as tracked vehicles or remotely piloted vehicles, and used to service large numbers of robots operating in the local area. The robots might then be able to get by with a less capable and less expensive sensor. Navigation and pilotage sensors will also have to operate in adverse conditions, except that they will not have to have the long range of the threat sensors. (To differentiate, navigation is the planning of routes based on limited a priori knowledge of the local terrain. Pilotage is the act of steering a course along that route and avoiding obstacles that are sensed along the way.) The level of sophistication required in the pilotage sensors is directly relatedto the amount and accuracy of the a priori terrain data. On-board communications must be secure and jam-resistant. Communications links will probably be established between the robot and its command center, other robots assisting in the same mission, and other sensor platforms. Selecting the Platform The robotic platform depicted here is an eight-wheeled trailing-arm configuration. Its ground contact rivals that of track-type platforms but fails more gracefully. For instance, when a tracked vehicle hits a mine, the crew has no recourse but to exit the vehicle and short-track it in order to get moving again. Independently articulated trailing-arm links can raise a damaged wheel assembly out of the way without human assistance. Direct protection for the platform is a thin skin of armor plate designed to protect against small arms fire and artillery shell fragments. Direct protection against larger caliber weapons would require ponderous special armors and severely compromise mobility. Thus, protection from such weapons will have to be achieved through cover and concealment, which creates a size-vs.- mobility dilemma for the platform designer. The platform must be able to keep pace with and support the parent unit during cross-country movements and over rough terrain. This is to some extent facilitated by larger, more powerful chassis configurations. However, as vehicle size increases, positions offering adequate concealment become farther apart, thus extending exposure time. Platform designers must also closely examine the questions of reliability, availability, maintainability and durability. For example, battlefield robots must be able to operate for up to 72 hours between "pit stops." Even then, they may have to service themselves from fuel and ammunition modules that are air-dropped at some general location; the robots will have to find the modules and gain access to their contents. Also, combat vehicle crews usually spend a great deal of time on operator maintenance -- oiling, checking and adjusting at each rest halt. However, battlefield robots may not come in contact with their operators for days or even weeks at a time; thus, they will have to be designed for either self-maintenance or no maintenance at all. They will also need self-diagnostic equipment to prevent them from blindly attempting to execute a mission of which they are no longer capable. Guidance and Control The computerized guidance and control group is responsible for the robot's higher-order mission and route planning, and lower-order component control. Higher-order mission and route planning is performed primarily by artificial intelligence or symbolic processing modules, which in turn access more conventional numeric processing modules as needed. For example, the robot may require an unimpeded field of view of some particular piece of terrain for an ambush mission. Symbolic processing might conclude that the military crest of an adjacent ridgeline would be a likely place to search. Ray tracing from each candidate position to the objective terrain via conventional numeric processing of digital terrain data could be used to select the final position. The technology required to accomplish this higher-order planning in real time is still in the developmental stages, while lower-order component control is now commonplace. The robot receives its mission-level orders from human operators through the man-machine interface group; in turn, the robot must be able to report on mission status, reconnaissance, surveillance and logistics. It is important that this man-machine interface be easy to use, so that a single, non-dedicated operator can handle at least a squad or platoon of autonomous robots. To be worthwhile, robotic solutions to battlefield problems must not incur an additional manpower penalty. Hence, the commander of the robot unit will probably also be the commander of the supported unit -- a company commander or platoon leader, who already has plenty to do without leading a robot through its motions. Morever the robotic force cannot be dependent on precise, individualized instructions. Each unit must be able to receive its assignment and act upon it without detailed human inputs. Therefore, a properly designed man-machine interface is a critical requirement. Key Technologies This description of a battlefield robot concept suggests that many technologies are essential to its success. However, two technologies appear to be most essential to the overall feasibility of the concept: real-time artificial intelligence with embedded computer hardware and software; and automatic target detection, classification and engagement. Artificial intelligence theoreticians suggest that image understanding will be essential to a robot's vision system if it is to negotiate cross-country terrain autonomously. Image understanding would have to be done in real time to be of value; according to estimates, this would require at least a 6,500-rule knowledge base. Manipulating such a large knowledge base in real time would require approximately 100 billion operations per second of computer processing power, which represents a 100-fold increase over present computer processing capability. This requirement is good justification for the very high-speed integrated circuit (VHSIC) and massive parallelism programs that, although progressing, are not expected to pay off for several years. If one accepts all of these technical requirements, it is easy to conclude that autonomous robots will not be possible until VHSIC and massive parallelism products become operational. The flaw in this logic, however, is that image understanding is not essential to autonomous cross- country navigation. The ability to map local terrain in three dimensions is certainly essential, but 3- D mapping is not the same as image understanding. In other words, a robot does not have to know the difference between a rock and a tree in order to go around them. For example, ODEX, a six- legged walking robot developed by Odetics Inc., can autonomously negotiate a randomly dimensioned staircase without understanding the concept of "staircase." Artificial intelligence is best applied to functions like mission and routeplanning. The knowledge base required for these functions to achieve practical utility on the battlefield is far smaller than the 6,500-rule base needed for image understanding. Unlike many application areas, military tactics are relatively simple and straightforward: the smaller the knowledge base, the faster it may be manipulated, and the easier real-time performance is to achieve. The artificial intelligence modules will be expected to access more conventional numericid modules. Each type of module -- AI and numerical -- may be thought of as symbiotically reducing the search space of the other for efficient tactical problem solving. Recent advances in sensor technology have been particularly significant in the area of target classification, which can now be performed more easily with classical signature analysis than with AI-based 2-D image analysis. Brassboard versions of millimeter-wave ground surveillance radars are able to distinguish between vehicle models (an M60 tank versus a T64 tank, for example), and can track multiple targets with sub-mil azimuth accuracy and less than 10-meter range accuracy. These radars will be compact enough for mobile platform applications and will be capable of real- time performance. Addressing Program Issues Four major program issues must be resolved before this technology can be considered viable for combat applications. First, there has been a pronounced lack of funding support. Specifically, Army program element 63758 (Robotic Vehicle Demonstrator) has not been funded for two years. This program has been criticized because its evolutionary approach to autonomous operation via tele-operation is not sensitive to the near-term threat facing the user community. Perhaps this criticism carries a message for the developer community as well: must every new idea be developed as an iteration to a previous idea? Many advanced weapons systems have been developed this way, but the bottom line on their technical balance sheets suggests that these systems are not much more advanced than their predecessors. Second, a better balance must be struck between base technology and applied technology efforts. Currently, most applied technology efforts are held up until the requisite base technology programs yield their most advanced products. However, this advanced technology may not be essential to the development of autonomous robots that are of practical utility on the battlefield. The existing threat poses a clear and present danger that will only become worse in the near future. In the meantime, near-term tactical utility could be achieved through applied technology and innovative use of existing and developmental components. Third, and closely related to the need for a better technical balance, is the necessity for wide acceptance of the program's technical feasibility. In order for this program to succeed, we must have more faith in our industrial ingenuity and stop listening so closely to the theoreticians. Finally, we must start paving the way for eventual NATO acceptance of battlefield robotics. Without it, autonomous robots will not be available to us and our allies on the key battlefields of future conflicts. Mission autonomous battlefield robots are operationally and technically feasible in the near future. The user community has said it needs them if it is to achieve victory against present odds. It is therefore necessary for both government and industry to take another look at the technical and programmatical issues and to find more efficient ways to resolve them.