Processing Technologies Give Robots The Upper Hand
Autonomous vehicles are poised to change all aspects of military and civilian life.
The science-fiction image of robot warriors engaging in decisive battlefield conflicts may be closer to reality with the development of new robotic technologies. Researchers are sending autonomous vehicles up stairs, through fields and across a nation as they work toward mobile machines that can learn new behaviors while operating independently of human control.
Land, sea and air may witness machines performing human—or even superhuman—tasks vital to battlespace supremacy. Underwater autonomous uncrewed vehicles could search for floating or seabed mines. Unmanned combat aerial vehicles might lead strikes against heavily defended ground targets such as airfields. And, autonomous armored vehicles could form the advance wave of ground force attacks.
Robots have applications in all the military services but especially in the U.S. Army. The Army’s drive toward lighter, more mobile, rapid deployment forces may require robotic vehicles to meet weight limitations and low-profile requirements. Removing people from some types of vehicles allows designers to build them smaller and lighter. This becomes even more relevant with vehicles that sport thinner armor to save weight. The potential increased threat to human operators becomes moot when humans are no longer required in the vehicles.
The technologies needed for these robotic devices are well-known and understood by researchers pursuing them in government and academia. Dr. Clinton W. Kelly III, senior vice president, advanced technology programs at Science Applications International Corporation (SAIC), McLean, Virginia, founded a robotics program at the Defense Advanced Research Projects Agency (DARPA) in the 1980s. Kelly’s company recently was awarded a DARPA contract for PerceptOR, a program to develop advanced prototype perception systems for unmanned ground vehicles for the Army’s future combat systems effort. He relates that other scientists have advanced the robotic state of the art through a host of diverse, yet complementary, research programs.
“We now are able to produce systems that can operate in a sufficiently diverse number of environments to begin to do useful things in both military and civil contexts,” Kelly declares. “Until recently, we couldn’t do very much at all.”
For example, mobile robots in 1970 could move at a speed of only 6 meters per hour. This would involve movement in 1-meter increments separated by 10-minute processing breaks, and after a while many outside conditions, such as sun angle and lighting, would change. This would affect the robot’s ability to traverse more than 15 meters.
Now, robotic vehicles can travel across land at more than 70 miles (112 kilometers) per hour on a well-marked road in daylight and good weather. This leap in capability has occurred in just 20 years. Under more difficult conditions such as ill-defined dirt trails, today’s autonomous vehicles still can travel about 20 miles per hour while avoiding obstacles.
Central to these advances is computational capability. Processor improvements have paved the way for robots to assume greater autonomy and to determine the correct course of action in less time. The 1970 robot processed about 0.5 million instructions per second, while current models feature 20 billion floating point operations per second.
Autonomous vehicles now can follow roads, traverse cross-country terrain and avoid obstacles. Kelly notes that these autonomous vehicles perform well on well-delineated roads under high-visibility daylight conditions. Adverse weather or darkness hinder performance. Some vehicles equipped with forward-looking infrared (FLIR) systems perform well at night, he notes. Intricate intersections can confound the system, as can a wet surface or a change from a paved to a gravel road. More complex tactical behaviors such as taking cover or concealing itself also remain elusive.
Researchers at the Robotics Institute of Carnegie Mellon University, Pittsburgh, have been adapting several types of vehicles for autonomous operation. Dr. Chuck Thorpe, acting director of the Robotics Institute, explains that its navigation laboratory has outfitted 11 vehicles—two high mobility multipurpose wheeled vehicles (HMMWVs), three passenger cars, two minivans, a step van, two full-sized city buses and a jeep—for autonomous driving and for driver assistance.
In 1995, Robotics Institute researchers sent an autonomous automobile from Washington, D.C., to San Diego. Only for about 72 miles was a human operator required to seize control of the vehicle, and this largely consisted of a nighttime stretch of the interstate highway system in Kansas that had been newly paved and lacked lane lines. Rush hour traffic also posed a problem when vehicles were so close together that the robot’s sensors could not follow road markings. Now, sensor and processor improvements may eliminate even those limitations, Thorpe believes.
Four years ago the institute built a robotic combine that harvested 100 acres of hay without human operation. Cameras and a global positioning system antenna provided the necessary guidance. Thorpe states that this device cut alfalfa faster than a human-guided combine—and it can operate 24 hours a day. Researchers designed two versions: one that traversed in a linear pattern for rectangular fields and another that reaped in circles for center pivot irrigation fields.
A smart helicopter can perform its own maneuvers as opposed to human onboard or remote joystick control. Similarly, a smart forklift can be driven by an operator or turned loose to follow a path and precisely scoop up and place pallets.
Potential conflict areas for robotic vehicles in the Army’s future combat systems involve operation in urban, forested and mountainous settings. Designing a robotic combat vehicle capable of operating in all three environments poses significant difficulties. These might encompass mechanical challenges, such as being able to operate in these three terrain types; sensing challenges, such as detecting soft sand, mud or swampland; and communications challenges, such as being able to maintain connectivity whether in a dense jungle, a deep valley or an urban environment.
Effective command and control amid a mix of men and machines also poses a problem. Robots must be able to react safely and intuitively to swarms of people such as dismounted soldiers. This is complicated by the normal human reactions that arise in a combat environment among both the foot soldiers and the robot operators. “It is really going to take much more work in human factors than the robotics community has done today,” Thorpe declares.
Thorpe offers that the biggest challenge facing robotics researchers is perception. “The toughest part about building almost any robot is having the robot see,” he emphasizes. While humans differentiate among varying ground surfaces such as dirt, grass and pavement, they also consider that a large tuft of grass may contain a hidden obstacle such as a rock, or that a paper bag blowing too slowly across a road may not be empty. Similarly, people can predict consequences of suddenly occurring events such as a ball bouncing into traffic from a playground.
Giving robots the same degree of perception goes beyond equipping them with visual sensors. Thorpe relates that incorporating other sensors such as radar, infrared, sonar and laser scanners can add vital information to the robot’s decision-making process.
One active program at the Robotics Institute involves multiple cooperating robots. Researchers there have built teams of robots to play soccer against other robot teams, which addresses issues such as collaboration, communication and competition against an opponent. Thorpe notes that the ability to field successful teams of cooperating robots in a well-defined nonlethal environment easily can be translated to other applications.
Another effort aims to build an autonomous lunar roving vehicle that travels fast enough to always stay on the sunlit side of the moon. This would permit long-term operation entirely under solar power. Researchers plan to test a version of this vehicle this summer north of the Arctic Circle in Canada, where daylight extends for several days.
Dr. Robin Murphy, associate professor of computer science and engineering at the University of South Florida in Tampa, is working on an innovative approach that uses one type of robot to transport another. Machines in this configuration are known as marsupial robots because a larger mobile machine carries several smaller robots much as a kangaroo carries its young in a pouch. This approach provides significant advantages for deploying microrobots that otherwise would have difficulty reaching a target site.
Small robots that are useful in surveillance, search and rescue or military operations in urban settings can deplete most of their battery power during travel to the intended operational arena. A larger robot can deliver the smaller units to the surveillance site, serve as a communications relay and even provide power replenishment.
Urban search and rescue situations, for example, often are very dangerous for both humans and search dogs for several hours after an accident. Similar threats confront people in regions contaminated by hazardous material spills. The marsupial robots can deliver their cybernetic searchers immediately. Studies have shown that this approach to emergency response is faster than either the mother or the children alone can achieve.
Murphy explains that the marsupial mother robot also can shelter the expensive microrovers from environmental hazards such as extreme temperature changes. The mother craft can retrieve its children after the smaller rovers have completed their mission, recharge them and transport them to another location for a new mission.
The marsupial robot concept emerged from work on sensor fusion and perception. Murphy explains that one aspect of perception focuses on how to get robots where they are needed. Under a concept known as coaching, a taller mother robot with a greater field of vision than its lower riding progeny uses this advantage to direct its children. Cooperative perception takes this one step further by having the mother robot generate a viewpoint that complements that of the smaller units.
The University of South Florida laboratory is working on two prototypes. One, developed by Real World Interface, a subsidiary of iRobot Corporation, is an all-terrain vehicle known as the RWI ATRV. It is slightly smaller than a mail cart and carries three small tracked vehicles that resemble tracked boxes with rectangular flippers. This propulsion method allows the small robots to climb stairs.
In addition to transporting these power-hungry climbers, the mother robot can perform advanced processing that otherwise might overburden the smaller units with both size and power demands. It also can serve as an intelligent controller that coordinates the three children as they move about in the field. This could involve fully autonomous distributed control, with the mother robot assigning responsibilities to the children individually or even redirecting one to assist another, or centralized control, in which all of the children report to the mother or to a remote human operator.
The marsupial children can serve as surrogate sensor systems for the mother robot if the larger unit suffers damage. The children also can scout ahead for the mother in a difficult environment.
Murphy relates that this concept is not limited to ground vehicles. Her laboratory is working with Aero Environment to incorporate micro air vehicles that would be launched from the marsupial mother robot.
Advances spawn additional technical challenges, and Murphy’s team faces several as it tries to develop ideal marsupial systems. Her laboratory has created a software road map that she describes as “pretty straightforward,” but robot hardware often is costly, specialized and cutting-edge complicated.
This is especially true in docking the children to the mother. These small units must position themselves precisely to fit into the mother’s carrying area, but this effort can be complicated if a small unit suffers a breakdown just as it is attempting to redock with the parent unit. The two other children could be blocked from reuniting with their mother just as their batteries are running down. Murphy explains that researchers must determine how to deal with this potential situation. Possible solutions include having the mother marsupial cast aside the stricken robot to make room for the others.
While the power replenishment capability from the mother to the children may come in the next few years, near-term research focuses on enabling software and perception for the recoupling process. Other efforts aim at improving robust communications links between mother and children, especially in reducing power requirements and economizing data flow.
“We’re very pleased at the utility of this concept,” Murphy emphasizes. “To be able to make a contribution to search and rescue, to planetary exploration and to the defense of our country is very important.”
Murphy’s marsupials might be incorporated in urban warfare settings, where the mother carries the children up to a building into which they enter and scout all levels. Human controllers could set the degree of autonomy.
The sky literally might be the limit for the marsupials. These robot craft could represent the next generation of interplanetary explorers. A mother rover ranging across the surface of Mars, for example, could dispatch more mobile or specialized probes to closely examine any curious features. These probes also could return soil samples to the mother robot for on-site chemical analysis. Autonomy would be vital in these applications because of the long time lag for signals to travel between Mars and Earth.
Thorpe looks ahead to disappearing robots. In this concept, robots would become unobtrusive in the same manner that computers have become ubiquitous as embedded processors in everyday devices. Automobiles, for example, are becoming smarter as they are equipped with infrared cameras and collision avoidance radars. The trend is toward smart equipment that will have embedded robotics technologies, such as smart suitcases that follow their owners through an airport.
Several technological hurdles remain before mobile robots can attain their full potential. Kelly allows that research has brought robots close to the point where they can learn. For example, software can provide an autonomous vehicle with several behaviors, but these are constrained by the designers’ imaginations. On the other hand, a system instilled with values can judge the potential effectiveness of a particular action. Various lessons learned during the vehicle’s operational time can be applied to this learned behavior. A semi-autonomous vehicle might even learn from a human operator, which over time would reduce the amount of guidance or control needed from the human.
The ability to evolve totally new behaviors for unexpected circumstances is key to robotic development. This represents an advance over mere learning in that the robot would be responding sensibly to a new situation on its own. Research into evolutionary computing is leading toward this development, Kelly notes. One small robot, known as Nomad, over time evolved a behavior for negotiating its way around obstacles based on experience it gleaned from an obstacle course in a room. The next few years likely will see greatly increased research into enhancing the repertoire of behaviors, he adds.
Data fusion is another vital area of development. Currently, robots tend to incorporate data from a few sensors for specific functions. The machines largely do not incorporate data to produce a common picture of the environment that features all potentially available data. One goal is to empower the robot with a three-dimensional representation of its local and potential environments. This representation would far exceed the information provided by individual sensors. Its overall picture would be a backup resource when a single sensor failure would otherwise prove crippling to robots.
Another key area involves cooperative robotics. The use of unmanned combat aerial vehicles calls for employing multiple vehicles swarming over a target area. These vehicles would have to collaborate with each other during execution of a particular task, which would require self-organization.
One vital aspect that does not receive a lot of attention is independent validation and verification, Kelly maintains. Robots can be assessed in testing, but devices that incorporate greater autonomy may not necessarily stop at the programmed border. Keeping robots from doing something unwanted by their human creators, long a theme in speculative fiction, is a concern that people do not know how to address, Kelly warns. Currently, the only solution is to keep a human in the control loop, but this becomes unrealistic as robotic capabilities increase.
“At some point in time in this century, we’re going to have machines that are so capable that they will cause us to ask the question, ‘Is that machine thinking?’” Kelly predicts.