Publications

Lab Notes

ROBOTICS Let a Robot Do the Work	Posted November 2010
They’re not automatons, but these robots map, track, and locate things for you.

In the popular imagination, robots not only have personalities of their own, from the fearsome Terminator to the cuddly Wall-E, they also are capable of dealing with unpredictable situations in unfamiliar environments. In the real world, such autonomy is elusive; robots either follow pre-programmed instructions and sets of rules, or they’re controlled remotely by humans.

A new robotics program at Lincoln Laboratory aims to change that by developing systems that can carry out tasks with less human supervision. Through 2009 and 2010, the program was charged with seeing how existing robotics research might be applied to military needs.

"It seems like there's a lot of solid research going on in academia," says Eliahu Niewood, head of the Laboratory's Engineering Division, who is spearheading the robotics effort. "Very little of that is making it into Department of Defense applications. The research doesn't seem to be really coupled to what the warfighters need."

The issue, Niewood says, is that academic researchers tend to explore theories of how robotics should work, rather than building systems robust enough for actual use. "They take it as a solved problem, and from an academic perspective it might be, but too often it doesn't get used in the real world." Niewood hopes Lincoln Laboratory can help close the loop by taking algorithms that have already been developed and building them into usable systems.

The first stage of the project incorporated a mapping capability into a Packbot, the ruggedized machine from iRobot Corporation that the military uses for bomb detection and other tasks. The mapping subsystem, developed by Michael Boulet, relies on a ladar, an active optical sensor, to send out and measure the return of light pulses. With that technology, the robot is able to map out where walls are and where one corridor opens onto another. So far, the system maps in only two dimensions, but the researchers are trying to give it three-dimensional capability, allowing it, for example, to measure the heights of doorways.

"A big challenge is having the robot know where it is," Niewood says. Because it is indoors, it can't use a Global Positioning System (GPS) signal, so it relies on an odometer to figure out how far it has traveled and when it has returned to its starting point. However, the odometer is not accurate enough on its own, particularly for longer routes.

Boulet and Byron Stanley are also heading the "Patrol Leader" segment of the project, which focuses on in-theater robot uses. Patrols in Iraq often send a sensor-laden manned armored vehicle in front of the main patrol convoy to look for hidden bombs. As the sensors are only a few meters away from the "on-point" vehicle's cabin, it is quite possible that a triggered improvised explosive device (IED) would injure the driver. Remotely controlling an unmanned IED-detecting lead vehicle from an operator control station located in a following vehicle decreases the danger to the soldier. However, watching a moving scene on a screen in a vehicle that's moving differently can lead to motion sickness for the operators, as well as depriving them of the ability to keep an eye on the rest of their surroundings. By adding limited autonomy to the robot, both motion sickness and operator loading are reduced.

In response to the military's need for safer bomb detection, the team is designing a robot that can autonomously stay on the road 20 to 30 meters ahead of the manned convoy while moving at 10 to 15 miles per hour, instead of the current 2 mph. They need a robot that is not overly complex, with a robust, dependable behavior. "We want something that's intuitive and that can be used by the troops without their having to worry that it might not react reliably in every case," Stanley says. One potential future capability is to detect obstacles, whether positive, such as debris in the road, or negative, such as potholes.

The robot would be equipped with a combination of GPS and an inertial navigation system to have a more accurate sense of its location. Most of the computer processing capability would be on the following vehicle, where the more expensive equipment would be safer. The team is also working on communications between the robot and the vehicle. "Communication is relatively difficult in certain environments, especially if you're in an environment with jamming," Stanley says.

The mapping project was completed last year; the Patrol Leader project is currently in its early stages. Niewood would like to see an actual product by the end of next year.
Further out, and more investigative in nature, is the third arm of the project, cognitive robotics. Headed by Jerome Braun, this effort aims to give a robot the ability to enter an unfamiliar situation with an open-ended task and make decisions about what actions to take. The hope is that a robot could, for instance, be sent to the site of a collapsed building, navigate and operate on its own within the rubble, looking for survivors.

"Whatever technology we have today is not at the level of perception and understanding for the robot to make good decisions by itself," Braun says. "They typically require a large amount of human involvement."

Sending a robot into an unfamiliar or unpredictable environment—an urban battle zone, for instance—requires the robot to be able to perceive and understand the situation, not just see its surroundings, but comprehend what it's seeing. Sparking wires or a burst water main might obstruct the robot's intended path, but there might be another path or a better way to achieve its goal. The robot needs to be able to discover those possibilities, and incorporate them into its decision-making process, and act accordingly. Because every situation may be different, the robot has to be flexible enough to deal with many variables. The idea, Braun says, is to mimic aspects of human cognition to allow the robot to cope with tasks that are not predefined in situations that do not allow direct human involvement.

In fact, Braun is focusing on developing biomimetic machine-intelligence approaches, inspired by biological cognition models. Such approaches, he says, can bring artificial intelligence and machine cognition to the level needed for a truly autonomous robot operation. A novel cognitive machine-intelligence architecture—the main focus of the cognitive robotics effort—draws on concepts from cognitive science and neuroscience, Braun says. As the components of the architecture are developed, they will be tested on example tasks. An example task, he says, can at first be fairly simple as long as it poses a sufficient challenge to demonstrate the cognitive operation and its strengths.

Cognitive processing by machine is obviously an ambitious goal. The cognitive robotics effort, Braun says, seeks to develop the technology underpinnings of machine cognition, embarking on the road towards intelligent and truly autonomous robots.

Niewood says the overall robotics project involves seven or eight Lincoln Laboratory researchers, with some collaboration with the Computer Science and Artificial Intelligence Laboratory on the MIT campus, as well as researchers at Carnegie Mellon University. The Laboratory is funding a related project at Olin College, a small engineering school in Needham, Massachusetts, and will work with some students from Worcester Polytechnic Institute starting in the fall of 2010.

Niewood hopes to build up a community of people interested in bringing more sophisticated robotics into practical use. "This is a very attractive niche that the Laboratory could fit into."

top of page