Russian Wake-Up Call From Winter AV Trials

Global warming is doing its job on Moscow: in mid-December, there is very little frost or snow.

Yet, it’s a high latitude winter with long nights and darkness that streetlights cannot fully illuminate. This is the background for the driverless Winter City competition. Five autonomous vehicles presenting different technical paradigms must accomplish the 31-mile route imitating an urban environment in three hours in the presence of a squadron of human drivers and mannequins of passer-bys. No connectivity or distant guidance is allowed so it’s a pure self-driving experiment.

The spirit of the law

Yet in only the twentieth minute, the autonomous Ford Focus of the team of Moscow Automobile and Road Construction State Technical University (MADI) had ruined the organizers’ plan. The car stalled at an interjection where it could not be moved and blocked the traffic. Driverless rivals and regular cars formed what can be remembered as the first-ever multi-AV traffic congestion. Three hours later, the judges gave up attempts to fix the situation and restarted the race.

Humans have learned to cope with such issues by applying the letter and the spirit of the law, explained Andrey Vavilin, CEO at BaseTracK. Once they identify that the clogging automobile cannot move, they naturally form a temporary scheme of traffic control to round the vehicle, ignoring physical road markings. AVs cannot do this because, by the letter of the law, it is disobedience.

The lesson learned from the incident is that traffic organization will need a thorough revision before driverless fleets enter the streets, presumed Boris Ivanov, head of AV at Starline. Potential bottlenecks must spotted out and fixed.

Connected to group intelligence

It was also noticed that human drivers displayed anxiety in presence of AVs, said Ivanov, being unable to predict the unmanned. Humans use intuition profoundly to predict the neighboring vehicles’ actions based on a variety of subtle indicators such as the direction of gaze or a manner of steering. It makes a group of drivers ‘connected’ in the style of fish shoaling.

Conversely, driverless cars give no clues. “One of the conclusions is that AVs must inform neighboring drivers about their intentions,” said Ivanov. “After the race, I started to ask myself how I, as a driver, ‘sense’ that a car in the front is giving me way or is about to turn.” He proposed that a maneuver warning system must be introduced as an industry-wide standard.

Sharing a map

“Unlike other teams, we used LiDARS not only for detection of objects in blind spots but also for vehicle localization,” said Kseniya Shashkina, deputy director at Regional Science and Education Centre for Heavy Industry. “During the race, we could see how it improved stability of localization in dense urban areas where relying exclusively on navigational tools caused failures.”

Starline’s white Skoda Octavia showcased potential consequences of loosing location. Being fooled by satellite navigation issues, the self-driving system wrongly adjusted the route causing the car to bump into a concrete fence. It is unclear if the external sensors were red-flagging a threat of collision or why the decision-making module did not pay attention to them. Anyway, Starline’s Ivanov admitted that such accidents must be rendered impossible before the car enters public motorways. He said that after the race, the vehicle was re-programmed to stop if its location is unclear: “Our primary goal is to ensure safety of both the lost vehicle and the people and vehicles in the area.”

LiDARS can improve stability of localization only if dedicated digital road maps are provided. Winter City had instigated co-operation between the participants which Ivanov sees as the best result: “Particularly, we’re now planning to collaboratively develop digital maps.”

As autonomous cars are maturing, our understanding of the self-driving technology-related issues to be tackled are also changing. “AVs are doing well on smaller motorways and interjections,” said Andrey Vorobyev, deputy head of MADI ITS Competence Centre. “Driving in larger roads with complex signage and a lot of cars changing lanes simultaneously remains a difficult task. Also, work has to be done on multi-criteria prediction task.”

No team met the criteria necessary to win the cash prize worth $2.5M. In spite of that, all the four interviewed teams said they were satisfied with the results because autonomous vehicles were taking on the driving tasks with confidence unthinkable not so long ago, although the developers admitted they were not road-ready yet.

Leave a comment

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