Purdue News

December 18, 1992 Purdue Pete Paves Path To Robotics Future WEST LAFAYETTE, IND. – Robots soon may vacuum shag carpets, clean up hazardous chemicals, guard military bases or explore ravines on Mars, thanks to a navigational system developed at Purdue University. The Purdue system guides Peter, a 400-pound, five-foot-tall mobile robot with cameras for eyes and three wheels for feet, into the future of robotics research. "This is a one-of-a-kind robot that uses visual cues to navigate around obstacles," says Avi Kak, professor of electrical engineering at Purdue. "The vision-based reasoning and control system allows our robot to navigate indoors at an average speed of 25 to 30 feet per minute, using ordinary laboratory computing hardware. That's the fastest a mobile robot can move indoors using vision." Kak says other robots may move faster, but they are not as intelligent as the Purdue robot. "Robots that aren't vision-guided must move through a path of obstacles at least once before doing it again," Kak says. "The Purdue robot navigates through places it's never seen before because it has maps of the areas programmed into its computer brain, so it gets to its destination faster." Researchers at the University of Bundeswehr in Munich, Germany, have an outdoor vehicle, specially outfitted with visual cameras, which travels by itself at 60 miles per hour by visually tracking highway edges. But because indoor environments are more complex, Kak says, that vehicle wouldn't be of any use for indoor navigation, such as in factories or homes. "Most industrial mobile robots are not free to navigate," says Robert Cromwell, a Paoli, Ind., native who recently received a doctorate in electrical engineering from Purdue and stayed on to work with the robotics project. "They typically follow metal guides in the floor and might be used to deliver items from place to place, but they can't do anything outside that path. Our robot navigates and even avoids stationary and moving obstacles." Kak says in the next five to 10 years, autonomous, vision-guided robots like the one at Purdue could be specialized for particular industriai applications and also could help perform tasks that are dangerous or tedious for humans. For example, they could be used to clean up hazardous waste and chemical spills, to investigate radioactive environments, or to explore the terrain of other planets. "Even household robots that would do automatic vacuum cleaning without supervision and without bumping into anything are not far into the future," Kak says. For many years, researchers in sensor-based robotics have been striving for fast navigation using vision feedback. Computer simulation of human perception hasn't produced anything near what people can do, but Kak says fast, vision-guided navigation is a big step toward emulating human behavior. The work at Purdue has been funded by sources such as the U.S. Department of Defense and several Japanese electronics companies. The navigational system is detailed in the November issue of the research journal Computer Vision, Graphics and Image Processing -- Image Understanding. The Purdue team developed the system and tested it on the robot in the Robot Vision Laboratory, a research facility associated with the university's School of Electrical Engineering. Min Meng, a native of Dayton, Ohio, and a doctoral student in electrical engineering, explains how the robot works using this navigational system: "We've programmed the robot's brain, a computer, with a hallway model, or map, containing lines, edges and other features of the hallway -- the doors, the bulletin boards, corners," says Meng. "These are what we call landmarks. Television cameras on each side of the robot's 'head' provide the robot with visual images of the hallway. The information is digitized and compared to the model map of the hallway. After comparing what the robot 'sees' with what it expects to see, the computer plans a path for the robot to follow to its destination." The human brain works much the same way. The brain stores cognitive maps of areas it is familiar with and can call on these to mentally trace a route from point A to point B. "Once I input initial and final positions and tell it to start, the robot automatically determines where it is and begins to navigate through the hallway and around corners on its own, usually coming to within an inch of its final destination," says Meng. Another important feature of the system is that the robot also periodically checks to make sure it's in the correct position. Kak says this helps the robot move more accurately. "While it's moving in the hallway, there is a certain amount of uncertainty in the robot's position caused by factors such as the slippage of the wheels on the floor," says Meng. "If the robot's uncertainty gets too great, an uncertainty manager in the computer stops the robot and performs a selflocation to reorient the robot before moving on." Matt Carroll, a native of Huntington. Ind., and manager of the Robot Vision Laboratory, designed the ultrasonic collision-avoidance system on board the robot. "By firing ultrasonic beams, which we can't hear, and listening for echoes, the robot locates obstacles, usually students milling around the hallway, and navigates around them," says Carroll. "The Purdue robot also has microwave and infrared sensors which allow the robot to detect human intruders. Armed with these sensors, the robot can detect a person's breathing in a room, so it has great potential as a security sentinel." Purdue News Service: (765) 494-2096; purduenews@purdue.edu