The arrival of GPS receivers in cell phones led to a boom in location-based apps and services—everything from maps that show you where you are, to new kinds of social networking. But step inside a building and GPS often fails. Now a startup has technology that enables devices to know their position inside a building to within a few steps, and it hopes this could lead to a second wave of indoor location-aware services.

WiFiSLAM, which publically demonstrated its technology for the first time last week, enables a phone to work out its position by combining the “fingerprint” of nearby Wi-Fi networks with information taken from a device’s accelerometers and compass. The company was founded by students from Stanford University, with the aid of the university’s StartX accelerator program for startups.

Mobile devices already use Wi-Fi networks to refine outdoor GPS fixes by accessing databases maintained by companies including Skyhook and Google, created by driving around “sniffing” for wireless networks. However this technology can today only allow accuracy of 10 meters at best and is primarily aimed at outdoor use.

The technology is typically accurate to within a “couple of steps” of your current location, says Anand Atreya, cofounder of WiFiSLAM: “This accuracy will change how you interact with indoor environments.” The technology could aid with navigation inside large and complex buildings such as hospitals or airports, he says, adding that app developers will likely find more imaginative uses, too.

“Think about going to the supermarket,” says Atreya. “We can provide information relevant to the product right in front of you.” Another possibility is allowing users to find the nearest store clerk, as long as that person is also being tracked.

When a gadget using WiFiSLAM wants to know its location, it analyzes the signal strengths and unique IDs of all the Wi-Fi networks around it. That is matched against a reference data set for the area either accessed over the Internet, or stored on the device. The estimate of location can be sharpened if a gadget moves slightly, because WiFiSLAM’s algorithms can gather multiple fingerprints. Compass data and accelerometer signals capturing a person’s footsteps are also used to refine the accuracy of subsequent location fixes as a person moves around.

WiFiSLAM needs similar data to be gathered in advance inside a particular building before it can offer location fixes. A person running another special app must walk around a building a few times, entering every room at least once. Algorithms originally developed for robot navigation process the changing pattern of Wi-Fi fingerprints and footsteps to re-create the path the person covered. That trace is then manually associated with a map of the place so that WiFiSLAM can tell a user in that environment where they are.

Other technology that uses Wi-Fi to for location sensing relied on expensive additional equipment, says Atreya. “I could walk into your building and have Wi-FI location working within an hour,” he says, claiming this will allow WiFiSLAM to be rapidly adopted by many places.

Eladio Martin, a researcher at University of California, Berkeley, is part of a team developing another Wi-Fi-based location app that’s accurate to 1.5 meters. Like WiFiSLAM’s, Martin’s team uses Wi-Fi fingerprinting and needs no equipment other than a cell phone, although it is currently just an academic project.

“Public buildings and especially those related to health care are some of the main candidates for the implementation of this technology,” he says. Martin is not familiar with WiFiSLAM’s implementation, but says that academic work published by members of the company suggests they could reduce the computational load of calculating traces from Wi-Fi fingerprints, which would make the technology more scalable.

WiFiSLAM plans to deploy the technology in a number of hospitals—including Stanford hospital—as well as shopping malls. The technology will initially take the form of stand-alone apps for navigation, for example, an app provided by a particular mall. However, the technology could eventually be built into apps with more general mapping functions.