Russia’ StarLine has brought the maximum speed of its unmanned car to 60 km/hr. Even in the global market, that is quite a high speed for a UGV. The company does not plan to increase the speed any further and will for the time being focus on the car management algorithms improvement. According to Boris Ivanov, Director of the StarLine Smart Car, the project will be developed and involve open source community.
StarLine’s main line of business is car alarm and security systems, whereas a driverless car is a long-run project the company advances at its own expense without attracting outside investors. The total project costs are not disclosed, yet the amount is certainly in vicinity of several million dollars.
A driving system designed by StarLine is installed at a Škoda vehicle. The car can all by itself drive along a set route, follow the road surface marking, react to traffic lights and signs, identify and go around obstacles, yield way to pedestrians and maneuver as required. All that is accomplished through assistance of cameras, radars and an additionally installed lidar, allowing the UGV to detect objects in front of it (180˚). A radar can collect data even when it is raining or snowing, whereas a lidar can not do that. Yet a radar can not distinguish a moving object from a static one, that in turn can be done by a lidar. Those are totally different sensors, so the information they collect helps a smart car to operate in any conditions. Radars are rather stable and informative allowing objects detection at a distance of 100 to 150 meters and 45˚.
In some while, StarLine plans to have several lidars installed to ensure better vehicle performance. If required, stereoscopic cameras can be installed as well. Unlike a regular camera, a stereoscopic one permits to make an accurate measurement of the distance to an object. The entire data collected by the sensors is stored and analyzed in the car’s central computer for making up a driving chart. StarLine strives to transform the data collected into a correct and safe driving chart. At the moment a StarLine UGV can drive at 60 km/hr.
“We do not go any further now. We need to sort all issues out at the present speed”, Boris Ivanov says.
A global travel planner is a navigator which sets points a vehicle should pass. Navigators designed by, for instance, Yandex, are already used in trip computers of the cars sold in the Russian market. A navigator though has no information on the situation on the road and can not show where exactly a car is located, if there are any barriers, how pedestrians move, etc. That is a task for a local planner which gets data from sensors (cameras, radars and a lidar), identifies current circumstances, detects and classifies obstacles, uses the driving chart information, monitors UGV’s location and thus adjusts the route. A local planner is tasked with situations as follows: if there is an obstacle, a vehicle should pass it on its left; if there is no road on the left, the obstacle should be passed on its right; if there is no road on the right, a vehicle should stop. There is a database with standard solutions which are used in practical situations.
StarLine employs already existing distinction algorithms, computer vision and neural networks. Yet neural networks can only give understanding of what is in front of a vehicle. What is to be done about it is the question to be resolved by a UGV designer. It is dangerous to rely on a neural network taught at a driver’s performance since drivers can have not just good, but also bad driving habits and skills. That means, the company has to create movement algorithms allowing a UGV to resolve tasks on the basis of distinction algorithms. The problem is, how can such algorithms or neural networks be certified. In StarLine’s view, certification of separate algorithms can be an option.
By Natalia Kuznetsova