Navigation systems learn the way

  • 09-Nov-2009 03:55 EST
037.jpg

The new NavPrescience route-finding system indicates the paths the driver is likely to take based on what it has learned about the driver’s preferences and the current conditions—time of day, day of week, weather, traffic, and so forth. A possible hazard, which is denoted by the red exclamation mark, leads the system to suggest an alternative route, the blue line.

Every driver has certain favorite ways to get around town or cross-country depending on a variety of personal motivations. One might, for example, prefer to take a scenic highway, or, perhaps more likely, want to avoid traffic jams, toll roads, construction work, flooding streets, stadium crowds, or even an ex-spouse’s neighborhood. But when drivers are asked in research studies why they select one path over another, they often cannot clearly specify their reasons.

Knowing exactly why an individual would choose a certain road will not be necessary when smart route-finding software becomes available on GPS-based navigation systems. These new algorithms can reliably recommend map routes that bypass problem areas. Such technology could enable future intelligent driver assistants to better manage fuel efficiency, activate turn signals or directional headlights automatically when entering turns, engage automated glare-reduction systems, or provide other help depending on the location.

One effort to bring such location-based services (LBS) into the car is being mounted by a Pittsburgh-based start-up company named NavPrescience. It will soon offer a nav system-enhancement package that can autonomously gather knowledge of a driver’s route preferences over time and suggest appropriate alternatives.

“The new capability is intended to enhance a driver’s ability to safely and efficiently reach a destination on time with minimal stress,” said one of the system’s developers, Anind K. Dey, Associate Professor at the Human-Computer Interaction Institute of Carnegie Mellon University. The patent-pending NavPrescience system should be especially useful in “keeping older drivers in their comfort zone” by matching individual driving capabilities and profiles, such as avoiding interstates and unguarded left-hand turns, he said.

One of its underlying technologies, NavPrentice, is a routing system that leverages all available data that could influence path selection including time of day as well as real-time roadway information, traffic, and weather reports. NavPrentice then observes and, in effect, learns a motorist’s intentions, habits, and preferences to generate optimal routing solutions by emulating an individual’s patterns. A related subsystem, NavProphet, then accurately predicts the driver’s destination as well as favored routes and turns.

“We had this idea that we could learn people’s preferences for any task in situations where they couldn’t express themselves clearly,” Dey recalled. “For every choice you make, there are a lot of choices that you don’t make. So if you have enough examples, you can understand the ‘space’ of an individual’s choices and figure out the reasons underlying why they make a specific decision and not some other one.”

This kind of issue is known in the field of artificial intelligence and machine learning as an open-world problem. He offered another example of such a problem: an investigator could, for instance, predict a business traveler’s typical choices of airlines by asking directly but could tell almost as much about the traveler’s habits by observing them over time. Dey and his colleagues’ early work at CMU to apply these concepts to GPS navigation systems was supported by grants from the National Science Foundation.

The resulting driver-assistance software can reside either on a laptop (the in-car system) or on a server that enables interactive Web-based services such as rapid generation driving directions that can then be forwarded via mobile phone. The NavPrescience system is always on, but it interacts with the driver only if it discovers some sort of impediment along the standard pathways. Information delivery to the driver on the screen is simple: the darker the red line on the map, the more likely the route that the car is on. Along the way, the higher probability pathways are constantly updated to present the driver with options.

NavPrescience recently incorporated and has attracted some seed money funding from a Pittsburgh-area incubator called AlphaLab, which was established by Innovation Works, a venture capital group. Dey reported that the start-up firm is talking to potential commercial partners including GPS unit makers and resellers, car companies, and telematics services about incorporating the novel capability into their existing systems. The company hopes to ship its first product in the first quarter of 2010.

The NavPrescience system resembles prototype technology now under development by Andreas Winckler and his colleagues at BMW’s Munich research department. The new system, called Ilena, short for Intelligent Learning Navigation, can also guess where motorists plan on going and predict the route. It generates a list of potential destinations by monitoring and recording motorists’ driving habits and the locations that they visit frequently. After compiling the data and establishing a personal profile, Ilena can accurately deduce where the driver is heading. As the smart system becomes acquainted with its driver and common routes, it can even identify road gradients, curves, and braking points on a particular road.

In ongoing evaluations on a BMW 3 Series, Ilena has shown itself to be accurate about 80% of the time, a capability that BMW researchers claimed will only improve with time. The machine-learning system is also able to analyze its fuel stores and then optimize the vehicle’s performance for a particular destination, potentially offering a 5-10% boost in fuel economy, they estimated. A market-ready product based on Ilena could appear in three to five years.

Both smart navigation systems presage a new generation of sophisticated driver assistants that will introduce personal relevance to existing in-car information technologies by sniffing out telltale keys to an individual’s choices over time and emulating them. But as these evolving capabilities start to take ever greater advantage of outside information obtained from wireless networks, driver privacy issues are likely to arise when motorists’ personal information begins leaking (almost inevitably) into the wider world.

Share
HTML for Linking to Page
Page URL
Grade
Rate It
4.80 Avg. Rating

Read More Articles On

2016-09-06
Model ASUH miniature hybrid high-integrity pressure transducer from KA Sensors is designed for use in rugged environments from -65 to +300ºF (-54 to +149ºC) and vibration levels of more than 20 g.
2016-09-01
Time of flight (ToF) cameras are ready to let drivers control some of the many options of today’s infotainment systems with a mere wave of their hand. ToF-based systems can also monitor drivers to see if they’re drowsy or not watching roadways.
2016-09-20
While OEMs wait for NHTSA's V2X mandate they are discussing whether broad usage can be achieved without regulations.
2016-11-13
Lengthy automotive development and production cycles have long prevented automakers and startups from working together. While that’s changed a bit, many young companies still find it difficult to work with OEMs.

Related Items

Training / Education
2010-03-15
Article
2016-09-06
Training / Education
2007-03-01
Article
2016-11-15