Indoor Location's Upcoming SLAM Dunk

The beauty of GPS is that it works everywhere. With a GPS-enabled phone, applications ranging from navigation to picture geotagging, from nearby restaurant search to social network check-ins, can all work everywhere. That is, everywhere outside. Once we go inside, our phones simply remember where we were last observed.

There are over 170 companies developing technologies for indoor location positioning, enabling phones to track their locations inside malls, stores, offices, hospitals, airports and more. Not a week goes by without announcements being released of new deployments of indoor location technology. But virtually all the indoor location systems on the market have to be deployed and customized one site at a time. And deploying and customizing these systems takes a lot of time and effort.

A new technology brings the promise of changing all that, of enabling indoor location systems to work everywhere, or enabling them to be deployed in new much much more easily. SLAM stands for "simultaneous localization and mapping," basically a self-learning approach whereby a system learns about a site automatically while walking around it, and does its best to estimate locations from the start. SLAM has been around in the aeronautical and robot communities for years, and is now being implemented for mobile.

According to a new report from Grizzly Analytics, there are at least eleven companies working on bringing SLAM to market, ranging from the biggest mobile and technology vendors to the smallest start-up companies. The best known of these companies is Apple, who has been known to be working on SLAM for a while, based on their acquisition of start-up WiFiSLAM, but there are many more as well. Interestingly, Cambridge Silicon Radio (CSR) is developing strong SLAM technology, both in software and for integration into their chips. CSR was recently acquired by Qualcomm, makers of the IZAT chips that use radio signals for indoor location positioning. Adding CSR's SLAM to Qualcomm's radio-based positioning could mean that 2016's smartphones could do indoor location positioning anywhere, with millions of phones SLAMming their way through new sites.

To understand how SLAM works, consider the two most common indoor location technologies already on the market. The first is called fingerprinting. In fingerprinting, a phone or laptop is used to record the Wi-Fi and Bluetooth signals at hundreds of places throughout a site. The collection of signals at any place is called the "fingerprint" for that place, in that the exact collection of signals should be slightly different at every place. Generally speaking, signals should be stronger the closer the phone is to the source of the signal (the Wi-Fi access point or the Bluetooth beacon), but signals also vary based on what objects are interfering with the signal, walls that signals can bounce off of, and many more factors. Recording fingerprints in a database enables a system, in principle, to identify a location based on the set of signals observed.

There are, of course, a lot of complexities and challenges in implementing a fingerprinting system. But even the best and most sophisticated fingerprinting system requires that fingerprints be gathered every few meters around the site, which can take days for a large sized site.

Consider now another common technology for indoor location position, called sensor fusion. Sensor fusion basically uses sensors that are built into phones, such as accelerometers, gyroscopes, compasses and barometers, to sense the phone's movement and to track the changes in its location by precisely measuring its movement. In other words, if the phone starts out knowing that its at point A, and then measures that it moved one meter south, it should be easy to calculate the new location.

Motion sensing of this sort, however, is notoriously inaccurate. First, the sensors built into today's smartphones are not top-of-the-line professional sensors, they're much smaller and designed for simpler tasks. Second, small amounts of error tend to build up very quickly. If the measured direction of movement is off by just one degree from the actual direction, the location estimate would be very inaccurate after just ten minutes of walking around.

Enter mobile SLAM. The most common approach to mobile SLAM is to combine sensor fusion with fingerprinting. This type of Mobile SLAM system uses sensor fusion to track its location at well as possible when moving around a new site. It also records radio signal fingerprints as it goes. By combining the data from ten or more walks through the site, it can automatically put together a map of the site and a set of fingerprints for locations throughout the site. These fingerprints can then be used for indoor location positioning.

Clearly, there is a lot more to it than this simple description. If sensor fusion is very inaccurate, how can a SLAM system know where the phone is when it's gathering fingerprint data? There are a lot of technology details that different systems used to make this SLAM process work effectively. Some systems assume that the phones will occasionally walk near windows or open areas of a site and receive sporadic GPS signals that will clarify the phone's location. Some systems detect when the phone is at the same place twice, by recognizing the fingerprints, and then review all the movement data collected between the two times at the same location to clarify the path based on the ending point. Some systems detect when two different phones running SLAM are near each other, to clarify the location estimates of both. These and many other innovative approaches are being used to make SLAM work.

Other systems are taking a very different approach to SLAM, using device cameras to implement what I call Visual SLAM. In visual SLAM, real-time images from the phone's camera are analyzed to detect the phone's movement much more accurately than can be done using the phone's other sensors. Also, in addition to collecting radio signal fingerprints, visual SLAM systems collect images around the site, which can also be compared to determine where the phones are, and can be used later for more accurate location positioning. Visual SLAM, however, is often more power-hungry, and cannot be used for a long time without running down a phone's battery.

Bottom line, mobile SLAM systems present a vision of the future, in which phones can do indoor location positioning anywhere. For sites that have not yet been learned, the phones would use SLAM to learn them. For sites that have been learned already, simple fingerprint matching can be used for location positioning. Suppose, for example, that mobile SLAM was running on all iPhones. With the huge number of iPhones otu there, how long would it take before all popular indoor sites had been fingerprinted and could support indoor location positioning? The same would be true if any popular smartphone brand or application was SLAM-enabled.

Some mobile SLAM systems have a more modest goal: enabling sites to deploy indoor location systems with a fraction of the setup time and effort. Instead of laboriously carrying out the manual fingerprinting process, which can take days for some big sites, the site owners simply need to walk around the site for a while carrying SLAM-enabled phones. The SLAM system would take care of all the fingerprinting, much more easily than using today's manual process.

Whether it comes to market in a universal manner or by making site-specific systems easier to deploy, SLAM has the power to revolutionize the indoor location market. Who will get the SLAM dunk first?

No comments:

Post a Comment