Published on FORBES.COM
Link to publication
Russia’s capital city of Moscow will soon be home to an ambitious autonomous tram project, joining Germany and China as the leading testbeds for the technology.
Self-driving vehicles are not currently legal in Moscow or anywhere else.
Autonomous trains and trams are stepping stones to full self-driving vehicles, thanks to their defined routes, but retain some of the hardest challenges cars face: knowing when obstacles, people, or vehicles are in their paths. That’s because they use public space, and even though there may be fencing or other barriers in some locations, people, vehicles, and other obstacles are still real concerns.
That’s the challenge that Cognitive Technologies CEO Olga Uskova says her company is close to solving.
“Cameras … correctly recognize objects in 80 percent of cases [while] additional data from radar raises the detection accuracy to 99 percent and higher,” Uskova said in a statement. “The Cognitive Low Level Data Fusion technology that is developed by our engineers allows the computer vision model to efficiently use all the combined ‘raw’ data coming from cameras and radars. This integration of data from different devices makes it possible to fill the missing information for better understanding of the current road scene.”
Cognitive Technologies is also building a self-driving car, which I took a drive in last year, and has already launched a self-driving tractor for large farms.
The autonomous tram will start as an “intelligent assistant,” Cognitive Technologies says. With more testing, the software will eventually start operating the tram itself, but until legislation changes to enable full autonomy, an operator will remain in the cabin.
Even after then, passenger psychology might still demand a driver.
I asked Uskova for more details.
Koetsier: How much simpler is a tram than a full self-driving car?
Uskova:Well, of course moving along the rails at a relatively low speed somehow simplifies the whole task. But in this project we still haven’t searched for easy ways. And we didn’t set a requirement that the tram tracks must be fenced.
For such a big city as Moscow this kind of a desire may cost a lot. Therefore we tried to train our system with the most possible number of unexpected situations the tram may face on the track. Here is one bright example: its late night, poorly lit street, a drunken man decided to sit down on a rail and fell asleep …
So for this project we used the most advanced technologies for obstacles and living creatures’ detection.
Koetsier: In terms of technology, is this a stepping stone to your self-driving car, or a separate project?
Uskova:This is one of the options. A 100 percent safe self-driving car that can drive on any road in any weather conditions is the most difficult task in the field of AI development for autonomous vehicles. That’s why I would say that all other tasks and projects can be considered as derivative works of the main and the most important task.
Koetsier: Regulatory environments also matter. Is this a way for Moscow/Russia to start approving self-driving technology?
Uskova:Yes. This is the beginning of development of Moscow’s smart neural environment that will remove a lot of intermediaries from many vital processes. Self-driving tech will systematize and protect passenger and cargo transportation, it will make life in the city more environmentally friendly and safe. Big cities are choked with traffic jams. The lack of parking spaces is a source of constant stress for citizens as well.
We lose time and we lose health in these conditions, so we have no other option – we must change the situation. Our team started with trams and we look forward to trolley buses and other public transport.
Koetsier: You have some strong beliefs as a company about what vision tech autonomous vehicles use. Can you elaborate?
Uskova:For over twenty years we have been historically involved in the development of computer vision (technical vision). But along with standard approaches, since 2012 we intensively focus on Deep Neural Networks.
This is a fundamentally different mathematical apparatus that requires an original fresh look. Cognitive Technologies now has one of the strongest teams on neural networks in the world. We have actually created our own DNN class that is adapted specifically for the autonomous vehicles. Such a strong mathematical apparatus allowed us to move from the tasks of creating AI to the tasks of a new level – creating Artificial Intuition.
This provided us with a real breakthrough in perception accuracy of our systems and with an ability to confidently speak about the industrial implementation of our systems.
Koetsier: Any updates to your predictions on when we’ll see full self-driving in Russia?
Uskova:Frankly speaking, no. The formation process necessary for autonomous driving legislation is very slow. I think that the first countries that will finish their transition to self-driving will be United States and China.
Koetsier: Thank you for your time!