Russia Attempts Road Map for AVs

Russian developers of driverless vehicles, intelligent transport systems and geographic information systems agreed to develop a unified format of a AV-specific open-source digital road model.

The opening conference, held in Saint-Petersburg last year, had seen allied companies share knowledge and practical approaches and align their strategies. The idea of the co-operation had first come at the technological competition Winter City where AV developers could observe and compare the approaches and findings of each other. It turned out that each team had scanned and digitalized the physical track ahead of the actual rides, probing their own paradigms as long as a unified approach for the whole industry does not yet exist.

“It’s like we all suddenly met at a junction when coming from different directions,” said Vladimir Zotov, chief specialist at KB Panorama. The importance of the issue can be seen from the fact that a lack of a reliable digital road model was the primary reason why Innopolis University team performed poorly at the contest. While developing a good model is just a start, an easily foreshadowed industry-level issue is digitalizing tens of thousands of miles of public motorways, said Boris Ivanov, head of AV project at Starline. It is a challenge feasible only for the biggest companies like Google or Yandex.

However, those players do not share their deliverables with the rivals to benefit from a competitive advantage, thus forcing the community of the AV developers to a conclusion that an open-source database, available for everyone to use, would help to prevent monopolization and boost progress in the driverless industry.

HD road map databases do exist designed specifically for autonomous driving do exist, including the open-source ones such as OpenDRIVE but the concept of a digital road model is not equal to an HD-map, said Andrey Vorobyev, deputy head of MADI ITS Competence Centre. Instead, the high-precision map is just a basement layer while the model also includes other layers such as traffic plans, weather conditions, assistance to scene recognition and many more.

Real-time awareness

A structure of layers and sets of data can vary from one AV developer to another but the uppermost layer is often the one that contains real-time traffic information. Making the model a dynamic object changing upon second precision, it is an indispensable part for Level 4 AVs, said Andrey Vorobyev. This year, street tests in Moscow had revealed that, in certain dense traffic areas, AVs could exercise driving only when guided by the ITS.

Another concern sounded at the conference was that the regulators can introduce a national standard of digital road models without proper consideration to the needs of the developers in favor of the larger companies such as Yandex or Sberbank. The risk is not merely theoretical: Alexey Ermakov, engineer at KAMAZ, had said that the work on the national standard had been under way earlier but stopped after carmaker AvtoVAZ had quit the board of its authors in 2017. The developers are hoping now to get ahead of the regulators’ next attempt on the issue by offering a ready draft of the future standard.

Another complex issue to be solved is compatibility of the future digital road model format with tons of related standards and technologies. On one hand, some geometric data can be derived from the digital terrain models for road construction and GIS services as well as the datasets created for driving assistants such as Continental’s eHorizon. Some topology requirements are also described in ITS standards, albeit, according to Vladimir Zotov: “We have never before faced the necessity to create such accurate and detailed maps.”

On another hand, not one of those systems fully answers to the needs of driverless transport because of either unsatisfactory accuracy or a different set of data, said Nil Podozerov, research and development director at RIPAS: “They lack many of the elements necessary for self-driving motor vehicles.”

Still, most of the data is to be obtained through direct scanning. Many developers had studied the potential of satellite imagery and discarded it because of inadequate accuracy. Drone-based scanning had shown better results in tests, said Aydar Gabdullin, a research engineer at Innopolis University, although the conclusion shared by all experts was that using LiDARs mounted on automobiles was the best available solution.

In the coming months, the members of the alliance plan to establish a online discussion board where a unified terminology will be elaborated and the sentiments of the developers aggregated, to be followed by release a first draft format and testing it in practice at one of the test sites in Russia.

Leave a comment

Your email address will not be published. Required fields are marked *