In research on vehicular networks, one important open issue is how to estimate the proximity and (road) traffic patterns over time. We have simulators, notably SUMO, which take maps and describe us simulated routes throughout the given map. However, it is still rather unclear as to their accuracy compared to the real world, and the simulations are not yet perfect. Especially in lower-latency applications (e.g. safety), we still don't have a clear view of what is possible. Also, SUMO is oriented towards simulation of urban road networks.
A few times, I've stumbled on papers that use traces from real world vehicles as basis for their research (see attached example). Due to privacy issues, they are usually obtained from relatively public vehicles like taxies. The question is, how does this correspond to the real world? Can we assume that the frequency of contact indicated by such traces is realistic? Can we use them for assumptions about the formation of clusters (or lack there-of)?
Conference Paper Recognizing Exponential Inter-Contact Time in VANETs
In my modest opinion, they are not. Taxis are a very specific kind of vehicle and they only represent a small percentage of the total amount of vehicles in a city. However, their movements and metrics (e.g., their avg. speed) are directly influenced by the situation of the rest of the vehicles and can also reflect the current traffic state.
Anyway, I suppose that some researchers use these movement traces, since there are not other "real" movement traces of "all" the vehicles in a city... but the results obtained in terms of vehicular communications could be unrepresentative or erroneous.
That's also my "gut feeling", as they say, but I don't have any evidence to refute the claim. Unfortunately, doing a proper study on this topic done as part of my phd is really far out of scope. I think some collegues at the university of luxembourg/security and trust are working on this, although publications are upcoming as far as I'm aware.
From a practical perspective: do we have any good alternatives? I think there's some traces from dedicated tracks out there, but I'm not positive. If anyone has any input, that would be fantastic.
The problems are that people hail or SMS taxis for short distances, not long distances because they are costly. The distribution of trip lengths may be somewhat different. The start points for the taxis are taxi ranks, which are not always representative clusters.
Another problem is that taxis seek work all day, but this does not represent the profile of the average daily commuter. So, you should limit the taxi measurements to the peak period.
Still, I feel that peak period taxis measurements, used with care do represent a useful, attainable samples of what travellers with cars will do, i.e., go from home to work or shops.
I agree with the comments above. Taxi trajectories do not represent the vast majority of driving patterns as they are mostly used for leisure or business trips.
Have look at the papers of my colleagues from TUM. They are looking at modeling spatial mobility patterns for households.
https://www.bgu.tum.de/msm/forschung/modellierung/traveldemand/
- Badges
- Science topic
- Similar topics
- Philosophy
- Moral Philosophy
- Free Will