The 3D algorithms of artificial intelligence

Robots could soon be capable of real-time autonomous navigation by creating and continuously updating an internal three-dimensional map.

Indeed, researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have designed such a system based on advanced algorithms and a low-cost interactive camera like the one found in Microsoft’s Kinect. 

To explore unknown environments, robots need to map in real-time as they move around – for example – estimating the distance between themselves and nearby wall, while planning a route around any obstacles.

Although a significant amount of research has been devoted to coding one-off navigational maps for robots, they are incapable of adjusting to changes in their surroundings over time.

“[For example], if you see objects that were not there previously, it is difficult for a robot to incorporate that into its map,” explained CSAIL researcher Maurice Fallon.

As such, MIT researchers have come up with a new approach, based on a technique dubbed Simultaneous Localization and Mapping (SLAM,) which allows robots to constantly update a map as they learn new information over time.

 The team has already tested SLAM on robots equipped with expensive laser-scanners, but recently demonstrated how a robot can locate itself in such a map with just a low-cost Kinect-like camera.

Essentially, as the robot travels through an unexplored area, the Kinect sensor’s visible-light video camera and infrared depth sensor scan the surroundings, building up a 3-D model of the walls of the room and the objects within it. Then, when the robot passes through the same area again, the system compares the features of the new image it has created – including details such as the edges of walls – with all the previous images it has taken until it finds a match.

Concurrently, the system estimates the robot’s motion via on-board sensors that measure the distance its wheels have rotated. By combining the visual information with motion data, it determines where within the building the robot is positioned. Meaning, combining the two sources of information allows the system to eliminate errors it might encounter if the robot relied solely on its sensors.
 Once SLAM is certain of its location, any new features that have appeared since the previous picture was taken can be incorporated into the map by combining the old and new images of the scene.

As noted above, the team recently tested the new system on a robotic wheelchair, a PR2 robot developed by Willow Garage in Menlo Park, Calif., and in a portable sensor suite worn by a human volunteer. They determined it was capable of locating itself within a 3D map of its surroundings while traveling at up to 1.5 meters per second.

Ultimately, the algorithm could allow robots to travel around office or hospital buildings, planning their own routes with little or no input from humans. 

“There are also a lot of military applications, like mapping a bunker or cave network to enable a quick exit or re-entry when needed… Or a HazMat team could enter a biological or chemical weapons site and quickly map it on foot, while marking any hazardous spots or objects for handling by a remediation team coming later. These teams wear so much equipment that time is of the essence, making efficient mapping and navigation critical,” Fallon added.