Anyone familiar with the travelling salesman problem will know that computers are not generally very good at route optimisation. This is not for want of processing horsepower, but because routing problems can be challenging to solve algorithmically.
A to B route-finding is a lot simpler, as the existence of my TomTom satnav system proves. However, the algorithms and heuristics employed in this kind of A-to-B planning are not very good at providing useful alternatives to the optimal path, for those occasions when the optimal path happens to have been dug up by the council.
At a Royal Institute of Navigation seminar this week I heard Alan Jones of a startup called Camvit (Cambridge Vehicle Information Technology) present his firm’s work on this real-world problem. Camvit has developed software capable of finding a range of good routes between A and B, rather than one optimised route.
As Jones explained, humans planning a journey - with an old-fashioned paper map - will tend to look first at the available trunk routes, selecting a few options and ignoring the fine detail until the major routes have been selected. They will thus tend to chose from a range of pretty-good routes, favouring known roads and avoiding known problem areas.
Jones cited the example of travelling from Cambridge to Manchester, where there are five distinct routes with very similar travel times, variously involving the A1, M1, M6, M62 or A1(M).
Satnav software, the AA’s online service or the likes of Google Maps will typically provide the user with just one route. Coarse adjustments are then typically possible, allowing the user to choose a route that goes via a particular point, or avoids particular roads, or goes via the shortest route, or avoids motorways. But these approaches tend to be very poor at providing a spectrum of routes with reasonable travel times.
Other adjustments tend not to help either. For a long journey, displaying the second, third, fourth and fifth best routes will typically create variation only in the fine detail of the route, at the local roads level near to the ends of the journey.
Altering the weighting factors (such as favouring distance over time) will also not necessarily throw up alternate routes. Jones noted that where a northern route is, say, 300km long, and takes exactly three hours to drive, an adjustment to the weighting will never find a southern route that is 301km long and takes three hours and one minute, even if there is a nice pub on the way.
After a lot of work, Camvit has come up with a clever approach called Choice Routes that it says is quite good at finding the kind of routes humans would spot.
It works by pre-calculating a vast set of conventional, optimised routes between points on the map. This process creates a unique, many-branched tree - sprouting out in all directions - for every point of interest on the map.
To calculate routes between two particular points such as Cambridge and Manchester, the software first compares the pre-calculated Cambridge tree with the prepared Manchester tree, and removes all routes that the two trees don’t have in common. Given that the branches of each tree will tend to be roughly radial, with the town of interest at the centre, typically there will be very few branches of any length that the two trees have in common.
“A surprising thing happens,” Jones said. “We find that there are a few long chains of marked roads, and the rest is noise.” These few points of obvious overlap are then used as seeds, forming the basis of the detailed routes that the software goes on to calculate. Play the above video to see how it works.
This turns out to create human-like solutions using a method that no human could employ - kind of like the exhaustive way in which computers compete with grand masters at chess.
There’s more to the choice of several good routes than meets the eye. Knowing that a range of routes is available also helps after you’ve set off on one particular route, because different routes will tend to have pieces in common. All five good routes from Cambridge to Manchester, for example, depart along the A14. The Choice Routes software is able to determine, therefore, these decision points: junctions at which the driver has the option to switch from one good route to another.
The driver can thus be alerted when these points crop up, choosing the route that seems best at the time. Or, with software connected to real-time traffic information, the Choice Routes software might give updated estimates of the best route in a much smarter way than current systems.
I also think there’s a big safety implication. From my own experience with a TomTom unit, I know that route switching is the time when the driver is most tempted to try to interact with the satnav while still driving. When leaving a motorway due to congestion, for example, TomTom will point you back onto the motorway until you can find a way to tell it not to, and there are seldom convenient places to stop. So anything that automates the business of realising that a driver wants to switch routes looks like a giant leap forward in safety to me.
Camvit aims to sell its system to the big navigation software suppliers. “If the techniques prove popular, we would hope to see Choice Routes become available as standard, or at least as a ‘give me choices’ button, in most journey planners,” Jones told me.
“The technology is protected by a patent application, and we are in advanced talks with several companies about their markets and pricing models,” Jones added. “We are always keen to hear from other companies that think the technique might be useful to them. Although our initial focus is on road journey planning for the individual driver, we are also keen to look at applications to public service dispatch of multiple vehicles, fleet management, traffic and travel planning of all sorts, and even systems as diverse as internet routing, pipework planning or integrated circuit layout.”
Recent Comments