While the general idea has been pursued for years, Sarnoff has achieved the one-meter accuracy milestone only in the past three months–an advance that will be published soon, says Kumar. It’s an important advance, says Frank Dellaert, a computer scientist at Georgia Tech. “This is significant,” he says. “The reason is that adding these velocities over time accumulates error, and getting this type of accuracy over such a distance means that the ‘visual odometry’ component of their system is very high quality.”
Kumar says the technology also allows users–whether soldiers, robots, or, eventually, drivers–to build accurate maps of where they have been, and also to communicate with one another to build a common picture of their relative locations.
Kurt Konolige, Agrawal’s research partner at SRI, of which Sarnoff is a subsidiary, says one goal is to reduce the computational horsepower needed to do such intensive processing of video images–something Kumar’s group is working on. But if the size and cost could be made low enough, he says, “you could also imagine small devices that people could wear as they moved around a city, say, or inside a large building, that would keep track of their position and guide them to locations.”
The technology, funded by the Office of Naval Research (ONR), is being tested by military units for use in urban combat. Dylan Schmorrow, the ONR program manager, says, “Sarnoff’s work is unique and important because their technology adds a relatively low cost method of making visual landmarks with ordinary cameras to allow other sensors to work more accurately.”
Kumar says that while the first priority is to deliver mature versions of the technology to Sarnoff’s military sponsors, the next step will be to try to produce a version that can work in the automotive industry. He says the technology has not yet been presented to car companies, but “we plan to do that.” He adds that the biggest ultimate commercial application would be to beef up car navigation systems.