MITRE engineers received a Technical Innovation Award and placed second overall in the Robot Rescue Competition at the Nineteenth National Conference on Artificial Intelligence (AAAI-04), held in July in San Jose, California.

MITRE works on several different robotics projects for our sponsors and we participate in such competitions to measure progress and validate engineering concepts developed in our research programs. For example, in a current MITRE-sponsored research project, we are exploring how to best use robot swarms. Previously, the researchers investigated robot platoon command and control.

Zach Eyler-Walker and Carl Burke prepare their robots for the competition.

MITRE artificial intelligence engineers, Carl Burke, Zach Eyler-Walker, and Dave Smith were the "support team" for the robots in this year's competition. "The goal of our research program is to develop technology to enable effective command and control of teams of robots," says Dave Smith. "Urban search and rescue (USAR) competitions like these provide a way to objectively evaluate our results and measure progress against other approaches from academic and industrial competitors from around the world."

Two major USAR competitions are held annually: at either the AAAI or the International Joint Conference on Artificial Intelligence (IJCAI) conference, and at the RoboCup event. In USAR competitions, robots explore a simulated disaster area modeled after a building with earthquake damage. Their job is to help out with the rescue.

Three Challenge Levels

The MITRE robot examines a "victim" in the Orange Zone.

The competition's arena, built to specifications developed by the National Institute of Standards and Technology, features three zones with differing levels of challenges to mobility and sensors. The Yellow Zone has a mixture of mirrored, opaque, and transparent surfaces, narrow passages, darkened areas, and obstructions such as curtains and blinds, overturned furniture, and a moderate amount of scattered debris. The Orange Zone includes more debris and stairs and a ramp to a second level with a hole in the floor. The Red Zone has multiple collapsed floors and an unstructured rubble pile with cinder blocks, dirt, tangled wire, hoses, and pipes.

In each zone, mannequins play the part of disaster victims. Competitors are scored on their ability to locate and gather as much intelligence as possible about these victims in 20 minutes. All victims have one or more identifiable attributes, including their form, motion, heat (from heating pads), sound (tapping or a recorded voice), and carbon dioxide emissions. Points are awarded based on the quality of information about each victim found, including the quality of a map indicating the victim's location.

Autonomous Trend

Each entrant can use any number of robots, which can range from fully teleoperated (remote controlled) to fully autonomous. "The most successful entrants in past competitions have used a single, highly mobile robot, remotely controlled by an operator with a joystick and a video monitor," says Smith. "However, MITRE is pursuing an approach using teams of robots with a greater level of autonomy. By enabling multiple robots to work in parallel, we hope to surpass the limitations imposed by teleoperated systems, which require at least one operator per robot."

In San Jose, MITRE's team consisted of three robots with a single operator. "Our Pioneer 3-DX robots use laser and sonar sensors for measuring the distance to obstacles," says Smith. "They also have a custom-built scanning pyroelectric sensor for detecting body heat, bumpers and infrared proximity sensors, and a color pan-tilt-zoom camera. They have an on-board computer and communicate via an 802.11b wireless network."

Sensor data is sent by the rescue robots back to the command console, where the data are fused to create a map showing the locations of the victims.

The ability of the robots to autonomously map the arena while searching for victims won the Technical Innovation Award for the MITRE team. "To give the operator the best possible situational awareness, our system combines sensor data from all robots and displays it on a single map," says Smith. "Multiple layers show the locations of obstacles and victims. We were able to make maps that you could hand to a firefighter or rescue worker. Some of the other teams may have been able to find more victims, but they didn't have the same capability to accurately place a victim on a map."

"We've been focusing on this mapping activity because it produces essential information that a human supervisor uses to monitor the status of a robot team," says Alan Christiansen, principal investigator of MITRE's Robot Swarms Project.

Using Lessons Learned

The MITRE team's second-place finish this year marked significant progress over the team's first competition at last July's RoboCup event in Padua, Italy. For example, the team discovered that, depending on the host organization, the size of the arena the robots compete in may be smaller than anticipated and that small, maneuverable robots may be needed.

And they discovered that to autonomously create maps based on the robots' sensor data, it's important to have accurate estimates of the routes traveled by each robot. The robots report their estimated locations based on counting the rotations of their drive wheels. A certain amount of inaccuracy in the robots' position estimates can be corrected using a technique called Simultaneous Mapping and Localization (SLAM), but too much error cannot be automatically corrected. The four-wheel-drive Pioneer 2-AT robots used last year skidded a bit when they turned, yielding mapping inaccuracies. The Pioneer 3-DX robots used this year have two drive wheels that are less likely to skid, enabling the SLAM software to create more accurate maps.

In addition, Eyler-Walker improved the code that brings in the sensor data and builds the map. "We're able to more efficiently build a map from the data points that are transmitted back to the command base," he explains.

"We also learned last year in Italy that we needed to prepare for bandwidth and interference problems," says Smith. "Many of the robots in Italy used the same 802.11b communications protocol as the MITRE team so there was a lot of interference with video signals, preventing enough data coming back to update the map.

"This year, we still had trouble communicating with the robots because of the distance between the operator's station and the robots. We solved the problem by buying some high-gain directional wireless antennas and doing some late-night hacking to get the cables to interface to our laptops. That gave us real-time updates for the map and the video. It made it much easier to understand what was going on in the arena."

Lessons for Military Sponsors

Participation in Robot Rescue competitions reinforced the MITRE team's belief that robotic systems in complex, unpredictable environments cannot yet be fully autonomous. "In the foreseeable future, our sponsors will require the ability to field teams of robots with human supervision for search and rescue, reconnaissance, and surveillance missions," says Smith.

Command and control of robot teams requires a combination of robust communications and the ability of robots to continue performing the mission during periods when communication is unavailable. A distributed communication architecture and increased robot autonomy may help achieve these goals.

Plans for the Future

Future work under our current Robot Swarms research project will involve expanding the team to include 10 or more smaller, less expensive robots. "To implement such a system, we need to develop new control and communication paradigms," says Smith. "A larger team of heterogeneous robots will need to be able to self-organize and cooperate to achieve mission goals."

The benefit of a swarm is that if two or three robots fail, the rest of the robots will be able to complete the mission. It's expected that a swarm will be able to cover a wider area more autonomously and provide more reliable command, control, and communication.

—by David Van Cleave and Dave Smith

Related Information

Articles and News

• Designing Robot Teams for Military Use
• Military Robots: Marching from the Lab to the Battlefield