Mobile Robot Laboratory Research Projects


AAAI Mobile Robot Competition


Ganymede, Io and Callisto: Champions of the "Clean up the office" task at the AAAI-94 Robot Competition


Faculty Advisor

Dr. Ronald C. Arkin.

Team

Acknowledgements

The project was supported by the CIMS/AT&T Intelligent Mechatronics Lab. The AAAI provided a generous grant to help the team travel to Seattle.

Movies

Click above to see more pictures, and videotape from the competition.

Description

Io, Ganymede and Callisto won the ``clean up the office'' event at the annual Mobile Robot Competition sponsored by the American Association for Artificial Intelligence (AAAI). The small robots, which look like miniature tanks, were programmed to clean up an office littered with soda cans, coffee cups and wads of paper. The group from Georgia Tech won the event by collecting more trash in ten minutes than any other team's robot.

Each year, teams of researchers and students from industry and academia bring their robots to the AAAI competition, which challenges them with a "real world" task. This year, the organizers offered two tasks; contestants could try either one or both. The first task involved navigating an office using an electronic map. The other was to clean up an office littered with soda cans, coffee cups, and crumpled paper. The contest was held in Seattle with teams from Stanford, Carnegie Mellon University (CMU), Lockheed Missiles and Space, the University of Chicago, the Colorado School of Mines, Simon Fraser University, the University of Bonn (Germany), Georgia Tech and others.

Georgia Tech's robots are programmed with a "reactive" system called motor schema-based control. Motor schemas can be thought of as low-level behaviors; "avoid obstacle" and "move to the goal" are examples. More complex behaviors are constructed by "adding" several motor schemas together. The robots sequence from one behavior to another as they accomplish their task of gathering trash.

Initially, a tiny color camera guides the robots to trash, which they grab with a specially-designed gripper. The robot's vision is sometimes fooled, so they may grab non-trash objects like table legs. But they can tell the difference since tables don't move when small robots try to carry them away; objects that do move are considered trash. Once a robot has a piece of trash in hand, it searches for a trash can, again using vision. After locating a trash can and moving to it, the robot drops the trash nearby.

Many people wonder why the robots are painted fluorescent green. Although it does make them a bit more flashy, there's another reason. Bright green is easy for the robots to see, so they use it to keep tabs on one another. They "cooperate" by moving away from each other, preventing interference and allowing them to cover a large area faster.

Even though the robots performed well, there was one problem: the competition trash cans were black, so the robots were programmed to move toward the darkest objects to deliver trash. Unfortunately, they confused dark shadows under tables with trash cans. The result was that they sometimes hid trash under furniture instead of throwing it away. Maybe that's just a sign that robots are becoming more human all the time.

The Georgia Tech team is a multi-disciplinary group of students and engineers led by Tucker Balch and advised by Professor Ron Arkin. Team members included, from the College of Computing: Tucker Balch, Gary Boone, Harold Forbes, Ray Hsu, Doug MacKenzie, and Juan Carlos Santamaria; from the School of Mechanical Engineering: Erik Blasch; from GTRI: Tom Collins and Dave Huggins; and Claudia Martinez, a visiting student from Mexico.