CHAMPAIGN, Ill. -- Robotics researcher Mark Spong at the University of Illinois has taught a robot to play air hockey. But Spong's work is not all play. His air-hockey-playing robot is a step toward the development of intelligent robots that could perform complicated tasks such as building a house, repairing a satellite or navigating through a network of pipes to inspect for corrosion.
"Our robot is a tool we use to illustrate theoretical problems in intelligent control and then explore them experimentally," said Spong, a professor of general engineering. "Our goal is to increase the performance of robotic systems by developing design techniques that combine mathematical methods from control theory with rule-based or heuristic methods of artificial intelligence."
Spong, along with colleagues Gerald DeJong, professor of computer science, and Seth Hutchinson, professor of electrical and computer engineering, designed and built the robot in the Coordinated Science Laboratory at the U. of I. "Building the robot was easy compared to actually teaching it to play air hockey," Spong said.
The environment in which the robot operates is highly uncertain and constantly changing, Spong said. The robot not only has to block the opponent's shots, it also must try to score goals. Therefore, the robot must continually sense its surroundings, interpret the data and react accordingly.
"In order to do that, we had to bridge the gap between artificial intelligence and control theory," Spong said. "For example, the robot's vision system includes a camera that tries to predict the motion of the puck. The robot then analyzes that information while planning its next action."
In the past, successful robotic applications have been limited to situations "where the environment is very structured and predictable -- like on an assembly line," Spong said. "One of the ultimate goals of robotic research is the development of robots that are capable of complex, autonomous behavior such as adapting to changes in their environments, planning and executing strategies to carry out tasks without human intervention, and learning from past experience to improve future performance."
Intelligent control is not possible without integrating the three areas of machine learning, computer vision and control theory, Spong said. "Our research combines techniques of explanation-based control with robust and adaptive nonlinear control, real-time sensing and sensor fusion, and motion planning."
Spong's air-hockey-playing robot, and the intelligent control research it embodies, may open broad new horizons for robotic applications. Many tasks, such as retrieving wreckage from the ocean floor or cleaning up hazardous wastes, are difficult or dangerous for humans. Spong envisions a time when such tasks can be carried out by robots.