Where Am I, Robot?
MIT Researchers Teach Machines To Make Maps
By BRAD PENISTON,CAMBRIDGE, Mass.
The red cylinder rushed through the labyrinthine campus building, spraying invisible pulses of laser and sonar energy as it rolled past the dean’s office, the nanotechnology lab and the war memorial. Wheels slipping on the slick terrazzo floor, John Leonard’s robot had the look of a fireplug that had somehow gotten horribly lost.
But few autonomous machines were better equipped to correct the condition of locational uncertainty. Leonard and his colleagues and students at the Massachusetts Institute of Technology (MIT) had endowed the robot with the groundbreaking ability to construct a map from sensor data — and correct it on the move.
This ability, called SLAM for simultaneous location and mapping, is among robotics’ Holy Grails, said Leonard, an associate professor with MIT’s Computer Science and Artificial Intelligence Laboratory.
“The better maps robots can make, the better they can do their missions,” he said Oct. 18 at an MIT-sponsored defense research conference here.
It is a trivial thing for a properly configured robot to dead-reckon its position by counting wheel revolutions and direction changes. But such reckoning is notoriously inaccurate, thanks to wheel slippage and other factors. Maps drawn from robot odometer readings go quickly awry.
Keep It Small
Leonard, along with another professor, Seth Teller, and graduate student Michael Bosse, realized that a machine attempting to build a single big map was generally doomed to fail. But a robot can build small maps with much less uncertainty.
The MIT scientists guessed that the robot could be programmed to fit these smaller maps together as it moved along, explored new areas and retraced its steps. A sophisticated pattern-matching algorithm called Atlas could even allow the robot to “close the loop” — to recognize when it was approaching an already-visited location from a new direction.
MIT Researchers Teach Machines To Make Maps
By BRAD PENISTON,CAMBRIDGE, Mass.
The red cylinder rushed through the labyrinthine campus building, spraying invisible pulses of laser and sonar energy as it rolled past the dean’s office, the nanotechnology lab and the war memorial. Wheels slipping on the slick terrazzo floor, John Leonard’s robot had the look of a fireplug that had somehow gotten horribly lost.
But few autonomous machines were better equipped to correct the condition of locational uncertainty. Leonard and his colleagues and students at the Massachusetts Institute of Technology (MIT) had endowed the robot with the groundbreaking ability to construct a map from sensor data — and correct it on the move.
This ability, called SLAM for simultaneous location and mapping, is among robotics’ Holy Grails, said Leonard, an associate professor with MIT’s Computer Science and Artificial Intelligence Laboratory.
“The better maps robots can make, the better they can do their missions,” he said Oct. 18 at an MIT-sponsored defense research conference here.
It is a trivial thing for a properly configured robot to dead-reckon its position by counting wheel revolutions and direction changes. But such reckoning is notoriously inaccurate, thanks to wheel slippage and other factors. Maps drawn from robot odometer readings go quickly awry.
Keep It Small
Leonard, along with another professor, Seth Teller, and graduate student Michael Bosse, realized that a machine attempting to build a single big map was generally doomed to fail. But a robot can build small maps with much less uncertainty.
The MIT scientists guessed that the robot could be programmed to fit these smaller maps together as it moved along, explored new areas and retraced its steps. A sophisticated pattern-matching algorithm called Atlas could even allow the robot to “close the loop” — to recognize when it was approaching an already-visited location from a new direction.
Comments