Serveur de calcul d'itinéraire
Lorsque vous utilisez l'application mobile Waze, les serveurs de calcul d'itinéraire utilisent des algorithmes spécifiques pour déterminer la meilleure route en fonction du moment où vous en faites la demande. Evidemment les détails du fonctionnement de ces algorithmes de calcul d'itinéraire ne sont pas publiés ouvertement car leur pertinence constituent une propriété intellectuelle de Waze. Toutefois, en se basant sur l'observation et surtout les informations révélées par le staff Waze, on peut circonscrire les grands principes qui régissent ces calculs d'itinéraires.
Ces informations sont sujettes à changement et certaines peuvent être incomplètes ou obsolète du fait de la mise à jour régulière des algorithmes par le staff Waze.
Calcul d'itinéraire autonome de l'application mobile
Par défaut, les calculs d'itinéraires sont effectués par des serveurs centraux dédié à cette tâche. Toutefois, lorsqu'il n'y a pas de connexion entre le terminal mobile et les serveurs centraux, alors l'application peut tenter, sur la base des informations dont elle dispose en mémoire (essentiellement les éléments de cartographie qu'elle dispose en cache), un calcul simpliste d'itinéraire, bien souvent non-optimal.
Serveurs de calcul d'itinéraire
Quand vous demandez à Waze de calculer un itinéraire, la requête est envoyée aux serveurs centraux de Waze. L'itinéraire calculé est ensuite renvoyé vers votre application mobile avec les informations temps-réel (trafic, dangers, accidents, contrôle radar,...) et vous est affichée.
Ces requêtes prennent en compte les paramètres que vous avez choisi sur votre application mobile. Dans le menu Navigation des Préférences, vous pouvez choisir
- l'itinéraire le plus rapide ou le plus court
- si l'itinéraire peut utiliser des routes à péage
- si l'itinéraire peut utiliser des chemins carrossables (ou non, ou seulement si le chemin n'est pas trop long)
- si l'itinéraire peut utiliser des autoroutes (sans considération du péage qui est traité séparément)
L'itinéraire le plus court signifie celui qui a la plus courte distance. La plupart des gens préfèrent l'itinéraire le plus rapide (100km d'autoroute est mieux que 90km de départementales)
Impact de la qualité et la complétude de la cartographie
Waze ne peut établir un itinéraire sur la base des informations de cartographie qu'il connait. Ainsi un itinéraire ne sera pas forcement le plus optimal s'il existe une meilleure route qui n'a pas été cartographié ou correctement connecté dans WME. Enfin, l'existence de petits segments (de moins de 5 mètres) provoque également des erreurs dans les calculs d'itinéraires qui peuvent aboutir à établir une route non-optimale.
Vitesses constatées en temps-réel
Le serveur de calcul d'itinéraire prend en compte les vitesses constatées en temps réel sur chaque segment impliqué dans le parcours (en se basant sur le remontées d'informations des applications mobiles des Wazers qui sont passés par là récemment) et les combine avec les historiques de vitesses moyennes par tranche de 10 à 15 minutes.
The time to pass through a segment is tracked separately for each route out of a segment. For example if a segment ends with a left and right turn (no straight let's say), then the routing engine isolates the time through the segment to turn left and the time through the segment to turn right. This detailed speed information is proprietary & is part of the data Waze sells/plans to sell and won't be displayed through the editor.
It is known that Waze uses real-time reports of current road speeds by preference over historical average road speeds. Waze also uses traffic congestion reports to reroute around slow traffic. It is assumed that the speed of any Wazer ahead of you on a stretch of road will be the speed used when your route is calculated.
As the number and density of Wazers grows, this real-time data takes on a greater importance. This emphasis is partly a reflection of Waze's original goal to create optimal commutes. As Waze has grown to be used as a more general purpose GPS navigation device over roads less traveled, the historical average road speeds becomes more important.
Since Waze uses the time you expect to be at a road segment to calculate the expected speed on that segment, it cannot use real time data when your trip time moves into a future time slot. So trips of more than 30 minutes (and on average more than 15 minutes) will include future time slots where current traffic data is not available when initially calculated. When you enter a new time slot, any real-time traffic data that is available causes your client to recalculate your ETA and reroute you at that point.
Waze knows the average speed of every confirmed road between you and your destination. The Waze server can calculate which list of roads to take to minimize the total travel time.
While every request is processed in real-time, by observation it appears that Waze caches some requested routes or major points. This means if it already knows the best route from B to C, and you ask for a route from A to C, it might just calculate the best route from A to B, once it checks that there isn't a better route bypassing B altogether. This does mean that when there is a Live map update some routes might be less than optimal for up to a day as routes are cached and recalculated.
While a complex calculation, calculating the optimal route is possible. The complexity arises over the "average speed" of the roads.
Changes in route due to different origin
Consider calculating a route from A to Z. The suggested route might be A to B to C to D to Z. Then calculate the route from B to Z. You might get a suggested route of B to C to E to Z. At first glance, this suggests one route is not optimal because the Origine changed.
We can guess at what might be happening. There might be a problem associated with the possible caching of routes. Or there might be a "timing window" effect. The different arrival time at C might change the best route to Z. Or there might be a fine tuning effect in the route, perhaps to avoid too many turns. So by dropping the segment A to B, the segments C to E to Z no longer exceed some threshold.
This effect can be observed while driving. If you recalculate a route to a destination while driving along an already calculated route to that same destination, the route can change.
Waze sometimes offers a new route to bypass heavy traffic which raises the question as to when and how Waze considers rerouting.
Changes in routes due to different route lengths
In addition, it is known that the Waze routing engine can't calculate every possible route for long distance routes, so it takes shortcuts for longer routes, potentially resulting in the route changing dramatically when lengthened slightly. According to Waze support, the section of a route more than 10 km from each endpoint is heavily weighted towards non-streets (i.e. "Primary Streets" or the various Highway types). It is believed that Waze servers simply ignore streets outside of the 10 km radius from the endpoints when doing its calculations to speed up the process.
Problems with average road speeds
If the average road speed is not correct, then the route will not be optimal. However, it is never recommended to delete segments in order to reset the average road speed for that segment. Waze uses the data from Wazers traveling through the segments to update the average speed. If you suspect Waze is not using certain segments along a route, there are a number of other reasons that can cause such a situation. Don't assume deleting the segment is the place to start.
The following are reasons why the routing engine might not use the average road speed.
Time of day variations
Consider a road that most Wazers drive at 5pm when the average speed is 12 mph. You choose your route at 10am when the road is clear and the average speed is 60mph. It is the best road to take, but Waze chooses a different route because it has no information about the speed at 10am and assumes the speeds is 12mph all day.
You can, of course, drive the road yourself, and Waze eventually learns the speed for that time. We do not know how long Waze holds the older data in computing the average speed. However, if Waze does discard old time information, you might not accumulate new time faster than it is discarded. It is also possible that your times are being ignored as being abnormal.
More Waze users would help fix this.
On a segment of a road, your average speed can be very different depending on what you do at the end of the segment. Traffic going straight through a traffic light might go very fast while traffic turning left might wait a very long time. A freeway exit lane might go fast while traffic continuing hits congestion.
- Seg4 to Jnct2 to Seg5
- Seg4 to Jnct2 to Seg6
Traffic building up on Seg4 that turns right to Seg6 does not affect the route timing for the traffic also using Seg4, but instead turning left to Seg5. For this reason it is important to keep long segment lengths before junctions if traffic congestion at the junction might affect exits differently.
To understand this problem better, consider if we add a short Seg8 between Seg7 and Jnct4. Let's say the traffic exiting Seg10 backs up all the way to Seg7 (easy enough, since Seg7 is short). Because Seg7 only has a single exiting segment (Seg8), the routing server is only able to collect a single average speed — it can no longer distinguish traffic by where it is going after Seg8. Now the through traffic going to Seg9 appears to Waze to slow down through Seg7, even though it doesn't in reality. At a minimum this causes an incorrect ETA for routing, and it might actually cause traffic to be rerouted unnecessarily, and less optimally, through another route. Hence if there is a chance that traffic that goes in different directions at a junction experiences different congestion, keep the segment before that junction long.
Note this data is not presented to users through the Waze Map Editor, but is only visible to the routing server.
Waze Map Editor changes
The average speed of a road can be dramatically altered by editing in Waze Map Editor (WME). Consider a 50m length of road driven in 36 seconds. An editor extends this length of road the full 10km length. WME now records this road as 10km long, but also driven in 36 seconds for an average road speed of 1000km/hour. Therefore be cautious of significantly changing the length of roads. Instead consider drawing new roads and linking them.
Traffic lights and stop signs
Waze does not record the location of traffic lights. While some GPS navigation offers guidance like "turn right at the next traffic light" the information is frequently incomplete, incorrect or outdated. The consensus view is that Waze should not record the location of traffic lights.
Waze does take traffic lights and stop signs into account by noting the effect they have on traffic speed. Consider a traffic light with long waiting times. The road segment leading to that traffic light will have a low average speed. If the average speed (based on the average waiting time) becomes low enough, a longer route that avoids the light becomes the preferred route. This has been observed in practice and is an example of emergent behaviour. Waze isn't programmed to avoid traffic lights but it does avoid slow roads; if the traffic lights make the road slow then Waze avoids them.
Some drivers regularly take longer routes — even winding through side streets — to avoid any stops or traffic lights. Waze has been known to suggest this, and also known to revert to waiting at lights when better average speed information is collected from the side streets. But note that this can be less than optimal due to the turn delays discussed above.
Outdated and abnormal road speeds
Road conditions change, construction work comes and goes, and average road speeds can change dramatically. One day you might be stuck behind a truck, and another day you might be crawling along the roads at 2 a.m. transporting your pet goldfish. Or your GPS might have a glitch and show you travelling at 1,000 mph.
In short, average speeds can change over time, and recorded times can be abnormal or just plain wrong—and can stay wrong for a very long time.
We can assume Waze is aware of this. There is some evidence that abnormal road speeds and old road speeds are discarded, or at least not used in calculating the average speeds of roads.
Average road speed "shrinking window"
If there are enough recorded speeds on a road, then Waze uses a shrinking window of speeds to better estimate the average speed at the time you are travelling on it. Waze uses the speed of each road segment (in both directions) in intervals as small as 30 minutes. So a two-way road might have up to 96 average road speeds. We can assume that when there are insufficient records for an individual time slot, a wider time range is takes up a full day. As the road is driven more, the time slot shrinks down to 30 minutes.
We can speculate that there is additional information used that is based upon the day of the week. There is some observational evidence that a road that is busy on weekdays and not used for routing is chosen as the optimal route on weekends.
At some point, Waze also needs to consider the months of the year as seasonal variations can affect the speed of the roads.
Waze uses the average road segment speed for the time slot that applies at the time you are expected to arrive at that road segment.
Because of this "time window", Waze suggests different routes at different times of day. This is dependent on how many times are recorded on the roads along the route. If they're all recorded at about the same time of day, then the time window does not help. If 1,000 times are recorded at about 5 p.m. and 2 times are recorded at 10 a.m., then your 10 a.m. average speed still is based mostly on times from around 5 p.m.
Routing algorithm refinements
It's easy to see how routing options can be implemented. For example, to minimize turns we can add a time penalty for fastest routing or a distance penalty for shortest routing, when calculating the fastest or shortest route.
For newly created roads, currently Waze applies a five-second time penalty for each junction through which a route passes. Once vehicles drive over that road and generate Waze traffic data, the penalty for these junctions is dropped.
What to do if you think the generated route is wrong
Firstly, use the option to generate alternative routes. This might give you some clues as to why Waze is offering that route.
Secondly, if you think there is a better route, check in WME that the roads are all connected along the route.
Thirdly, post a message detailing the problem route - origin, destination and a WME permalink to the Navigation forum. Other eyes will check it, and you might indeed find a flaw in the Waze routing algorithm. Fixing it might make it better for everyone.
Change of routing
When Waze receives notification of traffic conditions it uses the actual speed of roads on your route, rather than the average speed. This is based only on the automatic traffic condition reports - light, moderate, heavy traffic or complete standstill. Manual reports of traffic jams and accidents are for your information only and do not change routing.
It might be that even with the traffic reports, there is no better route and Waze cannot offer you one.
You are the driver
Waze can never see that the traffic light is green going straight, or know that today is a public holiday. It can offer you guidance as to what is the best route under average conditions. But you are the driver and you are in the best position to make the decision for today under today's conditions.
If everyone followed Waze directions and never drove on a new route, Waze might never learn that route is better. When Waze is recording your travels, every trip helps make Waze better for everyone. That includes when you think you know better. Sometimes you will be right. And sometimes you will be wrong. But it is better to find out you are wrong so you can choose the better route. And when you are right, all Waze users benefits by sharing in your knowledge.