VTI Technologies recently announced a project they worked on with Tampere University of Technology in Finland to improve navigation for both cars and pedestrians in environments where GPS may not be reliable. This can particularly be the case in dense urban settings where the GPS signal may be blocked or too distorted to be useful. With the growth in location-based services, even if you may not care what your exact position is, someone else does. Without GPS, dead reckoning is required to figure out where you are.
Dead reckoning refers to navigation where you have no clear points of reference along the way; you know your starting point, and based on your speed and direction, you calculate where you are at any time. The problem is that each of those calculations has some error and that, even if small, such errors accumulate over time and can make you think you’re somewhere you’re not. Before old-school gyroscopes, ships navigated by dead reckoning across the oceans, so you can imagine that, with very little in the way of reference points (except the stars, whose east-west position is confounded with time, making longitude hard to figure out), they might have been surprised when they finally caught sight of land.
Today GPS provides a golden reference point, but dead reckoning can be used in the gaps where GPS isn’t available.
I talked with VTI’s Ulf Meriheinä to understand better how this works. It’s a fusion of a number of technologies that come together to complement each other. The pieces of the dead-reckoning puzzle are:
– Access to speed information; this could be from the speedometer signal in a car or from a pedometer signal of some sort on a pedestrian (say on the chest-belt of a runner)
– A gyroscope
– A digital roadmap
– A simulation algorithm called “particle filtering”
The speedometer signal tells you how far you’ve gone in a straight line. There is an error of around 1% or less in this calculation. The longer you go without turning, the more linear error might accumulate.
Once you turn, the gyroscope detects that turn. So it can now help correct any accumulated linear error if it can determine accurately where you turned. And that’s where the map and simulations come in.
As you progress, your position is being correlated with the map. If you assume you can’t go off-road, then the number of places you might have turned becomes much smaller. In more technical terms, the probability density of your location gets compartmentalized, if you wish, to a few discrete possibilities.
This is where the simulation comes in. Particle filtering is a Monte Carlo variant where an overall probability density is partitioned into a discrete number of “particles.” As the simulation progresses, some of the particles are calculated as less and less likely and are filtered out, ultimately leaving one. Based on this, they system decides where you turned on the map.
Given that information, you can now reset your dead reckoning reference point to that value and restart your calculations as you continue forward. All previous errors disappear because the new location point isn’t based on your navigation calculations, but on deciding where on the map you are and picking the data from there.
VTI’s point with this is that you need a (or, more specifically, their) very accurate gyroscope for this to work so that those subtle changes in direction that we don’t notice, and which sometimes get us completely turned around, don’t escape detection and get this system all turned around.
Once GPS kicks back in, then you’re back onto that as a reference.