Next stop: conditionally-autonomous driving at SAE Level 3. For the control software needed for this to perfectly master critical driving situations, it needs to be trained thoroughly. And that is done using data gathered by sensors in test vehicles. Given that up to 40 terabytes are produced per day and car, the question arises: How can OEMs and suppliers handle such mountains of data?


Ein Sportwagen ist zu sehen, der über die Microsoft Cloud und einen PC verbunden ist.

Currently, the ambitions of the entire automotive industry are all moving in the same direction: taking conditional automation in vehicles at SAE Level 3 (SAE: Society of Automotive Engineers) to the point where it is ready for series production. At SAE Level 3, drivers only need to take control if the vehicle indicates an emergency situation.

Partly in order to design precisely this moment of handover to be as smooth as possible, the EU has announced a wide-ranging research project. The aim? To be able to trial SAE Level 3 functions in fleet tests, as far as possible error-free. The project is paying particular attention to motorway driving, traffic jam situations, city-centre driving and parking situations such as home-zone parking.

The special feature of the test fleets sent onto the roads for trials is that, in addition to the camera, lidar (light detection and ranging) and radar sensor assemblies fitted on and in the vehicles as standard, they are also carrying reference systems. These additional video cameras (four for recording the surroundings, three to record the interior) and lidar sensors produce a reference dataset, intended to show up possible failings in the series equipment a later stage, during the evaluation. Additionally, cameras in the interior film the face and feet of those driving, along with the instrument display in the vehicle, to record how people react at the moment control is handed back.


Result: the Cloud is keeping a grip on the data explosion

The result is that, on average, the test vehicles generate between eight and 40 terabytes per day and car. In such a mountain of data, it is barely possible to make out the critical driving situations in a meaningful way. However, these situations – defined by the EU, and numbering over 50 – are particularly important in moving to SAE Level 3. In order to master the mountains of data, the FEV Group (engineering services provider for all relevant OEMs and the EU project) has thought up a clever trick: their specialists developed an on-board mini data logger for the test vehicles, permanently monitoring the CAN-bus.

The logger continuously sends well over 200 signals (including information on distances to other objects, speed and the position determined using GPS) via mobile communications to a software solution developed by FEV and run on Microsoft Azure. This solution scans these signals for indications of critical driving situations – and marks any such moments in the flow of data. Subsequently, the raw data attaching to the time stamp can be extracted and used to train the control software under development.

For FEV, there was no question of setting up a dedicated Cloud infrastructure, for reasons of time and resources. Instead, the engineering specialists looked for a professionally-operated, key-ready environment that was compliant with all current data protection standards relevant for the sector. Azure scored highly, in part through functions such as IoT Hub or Stream Analytics, which are vital for the data logger project. What’s more, Azure also comes with a certification infrastructure, with the aid of which Azure only grants access to authorized end-devices. Additionally, using the certificates, the data recorded by the loggers can subsequently be correctly attributed to the respective vehicle.


Are you also often faced with a massive haystack of data, wondering how you can identify the needles hidden with it quickly and cost-efficiently? Then speak to us. Azure is bound to solve your requirements too.