Depends on what sort of analysis you want to do (does it have to involve machine learning or will you use stats/data mining/traditional traffic engineering algorithms). Where will the analysis happen? Will it happen on the vehicle IoT device or in a central server?
It shouldn’t be too hard to setup a pipeline that includes nupic (or other tools like scipy/sklearn) if they already have the data.
That said, I’ve got some telemetry data (from OBD (speed, rpm, MAF etc.) joined with GPS, this data will be private for confidentiality reasons). Eventually I’ll probably try to detect anomalous traffic flow in a spatiotemporal sense. ie. if we look at a sequence of telemetry data on a stretch of road(s):
Is the driving anomalous (in time and space)?
Does it indicate congestion?
Does it indicate an incident? (this is hard without more up to date data than I currently have)
Compare to existing ITS algorithms for congestion/incident detection for probe vehicles
There are two real world use cases for vehicle data. I saw these two below are computed
at the edge box (in the car). So algorithms are running in at the edge box computer. But it will be
a good practice applying HTM algorithms onto these. If someone need data set, we can provide by the way.
These two use cases above finally compose a driver score which will be used for a “Hall of Fame” in a country in which all driver scores are computed and published. In the end this list will be used in insurance domain while evaluating driver scores and insurance bills are computed according to these.
There is Ukrainian startup with a working solution in this sphere - they are using data from a mobile device in a car to detect quality of the road and apply this info to a map to build a better route.
I can share their contacts - perhaps you can find some common ground.