Free data for free research

Variety of situations and environmental conditions

The measured values are taken from four cameras, six LiDAR sensors, six radar, two inertial measurement units for determining the position in the surrounding space, and ultra-precise data from navigation satellites. “We also made sure that the measurements were taken in different weather and traffic conditions,” explains Fabian Kuttenreich, team leader for Perception in the Automation division at MAN. “This variety of situations and environmental conditions makes the “MAN TruckScenes” data set particularly valuable for research.”

And many teams from across the world obviously share this opinion – after all, the data set, amounting to several hundreds of gigabytes, has already been downloaded many times. “The high number of downloads has far exceeded our expectations,” says Leonie Wulf, who manages the Perception and Motion Planning department. “The data set has met with significant interest not only in Europe, but also in the USA, Asia and India.” And it’s no wonder: never before has a truck manufacturer made these kinds of measured values freely accessible. Not to mention that similar data sets from passenger car development are of little help for trucks. Trucks drive in different environments such as container terminals, usually have a very large trailer, need more sensors with a different set-up, and must also “see” significantly further in front due to their long braking distance.

Faster development through public data

With the published data, the international developer community can now develop better algorithms for highly automated driving – from which MAN in turn would also benefit. “We get free research for free data,” says Kuttenreich, getting straight to the point. So, the exchange of data with the open-source community speeds up development while also simplifying validation and encouraging standardisation.

How well the algorithms produced by the international teams work can be demonstrated in a challenge that started at the beginning of March. The task is to detect other road users and traffic signs in the “MAN TruckScenes” data as accurately as possible. MAN compares the submitted results against reference values and publishes a continuously updated ranking of the participating teams and their performance: MANTruckScenesDetectionChallenge – a Hugging Face Space by TruckScenes . “With the ‘MAN TruckScenes’ data set, we have closed a gap in research,” says Wulf. “Not only does the high level of interest prove how useful our data set is, but it also shows that MAN is highly esteemed within the developer community.”