[Routing] Using actuall collected travel-times as a metric

[Mark Williams]

Marcus Wolschon wrote:
> 
> Dear fellow routers,
> 
> I just finished a new plugin-type for TravelingSalesman, the
> IRoutingMetric. Thus I can exchange metrics at will.
> I am about to implement a very interesting metric
> http://travelingsales.wiki.sourceforge.net/FeedbackFastestRouteMetric
> that shall take into account actual collected travel-times
> to calculate the metrics for given routing-steps.
> 
> My questions to you are:
> 
> * Is anyone else interested in this?
>   We may share the data-pool.
> 
> * Where would one store such information?
>   I can provide my own web-service and
>   database-dumps to do this but we may also
>   store it in the map.
> 
> * What would be need to store here?
>   I was thinking about:
>   #number of samples
>   #average time
> 
>   and these 2 for all combination of the following cases:
>   * date and time
>   **weekday8-18
>   ** each hour between weekday19-20 and 6-7
>   ** weekend (not taking hollidays into account at all)
>   * type of car the user has chosen
>   ** car slow
>   ** car fast
>   ** car average
>   ** bicycle
>   ** ...
>   * metric the user wanted to optimize the route for
>   ** fastest (they would probably be faster drivers
>               for their type of car)
>   ** shortest
>   ** most fuel-efficient
>      (completely different style of driving at traffic-lights,
>       motorways, ...)
>   ** other
> 
Hi,

I went off & thought about this...

Would it not be more useful to have a general brief list of
driving-types associated per road, eg car/bike/walk, and actually have
your routing software adapt to _your_ style?

I thought perhaps this is perfect for a genetic algorithm, which could
encapsulate all the above list; this could then make its own
calculations for your usual speed off the lights etc, rather than make a
user decide who (s)he is beforehand.

Then the routing software can apply the rules to all routes passing the
basic test of connectivity and fine-tune the weighted timings based on
your habits; this gets around the need to acquire multiple tracings,
cuts down on amount of data needed, etc.

I would suggest that the accumulated data could be better used to train
a system like this than kept to use as static data?

Just my 2p as a user, I haven't tried actually doing this sort of thing
for some years now!

Mark