Contents
A team of students from MSTU. Bauman is working on a model of an unmanned racing electric car for international competitions. How does it work, how is the neural network trained to manage it, and what does the cloud have to do with it?
The Bauman Racing Team is engaged in the manufacture and assembly of car parts, while developing and testing an algorithm for training its control system inside the simulator. This is a virtual environment that simulates real racing conditions and an unmanned system. After the car is ready, the algorithm will be tested on it.
What will be the electric car-drone?
The team is developing the BRT8D car for participation in the international student engineering competition FormulaStudent of the FSD class (Formula Student Driverless). This is an unmanned vehicle with an electric power plant.
Under the terms of the competition, the car must be built in one year and comply with specific performance characteristics (performance characteristics). Here are the characteristics of the future BRT8D:
- Weight – about 280 kg, length – 2,7 m, width – 1,3 m, height – 1,1 m, clearance – 3 cm;
- Rear-wheel drive, maximum engine power – 95 hp, lithium-ion battery with a capacity of 11 kWh;
- Independent double wishbone suspension, electric power steering, pneumohydraulic dual-circuit braking system.
The finished car must fully meet the challenges of specific races. Unlike drones that drive on ordinary roads, it is guided by special color cones placed along the track. They help the car determine its position on the road and make maneuvers.
The electric car will be equipped with Nuvo industrial computer, Velodyne 3D laser lidar, stereo and mono cameras, special sensors and receivers of the navigation satellite system. It will operate using a power plant consisting of an EMRAX228 brushless electric motor, a Bamocar D3 inverter (motor power amplifier) and a battery.
In addition to lidars, the car will be equipped with accelerometers – special sensors that can determine the speed and acceleration of objects. These are found in modern smartphones. They help the drone determine where it is on the track and in which direction it needs to move.
Actuators in the brake and steering systems will allow the drone to move, turn and brake, simulating the movements of the arms and legs of a real driver.
Посмотреть эту публикацию в Instagram
The industrial computer inside the car runs on Linux, under which control algorithms are written. The main feature of the computer is increased reliability: it is better protected from damage, vehicle vibrations, overheating, dust and moisture. Inside the 9th generation Intel Core processor and graphics card, which are more powerful than in a conventional computer and consume less power. This is necessary so that the car’s battery does not run out in the process, as the computer performs complex calculations online.
The car is being developed from scratch and will be ready by August 2021. All parts are made to order in the laboratory or in production – this is also a condition of the competition. Sponsors provide something, some parts – for example, lidars, sensors, seat belts – are bought ready-made.
“The main difficulty lies in the fact that many things have to be done for the first time, without the ability to rely on any example – because it either does not exist yet, or it is in closed access. We are closed within the team and this is bad, because there should always be a look from an experienced person who can prompt in time, based on his experience, ”says Pavel Karasev, captain of the Formula Student team at Moscow State Technical University. N.E. Bauman.
Is it suitable for normal roads?
The BRT8D has a fourth level of unmanned capability. This means that it can drive autonomously, but only in the ideal conditions for which it was designed – on a special track with colored cones, without additional obstacles. There is no driver behind the wheel, but there is an operator who can stop the car remotely in emergency situations.
It is dangerous to release such a drone onto a regular road or race track for manned vehicles: it cannot recognize pedestrians or other sudden obstacles as well as Yandex or Google drones.
Why does a car need a neural network?
To drive a car, two convolutional neural networks are used as part of deep learning – that is, those that are responsible for pattern recognition. The first – YOLOv5 – looks for cones on the track and determines their color, and the second – KeyPoint – places special points that help determine the size of each cone and calculate the distance to it. So the system understands how fast the drone is moving and what acceleration it has.
Both neural networks are trained in the cloud using a special FSOCO dataset. It contains over 20 photos and videos marked up in a special way. This is necessary so that the drone can participate in different stages of the race: in different countries, with different weather conditions and different tracks.
What is the cloud for?
To quickly train the neural network, it was necessary to quickly increase the computing power. To save money on buying and maintaining their own equipment, the team decided to use Yandex.Cloud cloud technologies. Using this platform, the team was able to deploy a virtual machine with the necessary GPUs (video cards) in just half an hour.
“If we talk about the future, then, in my opinion, thanks to cloud platforms, heavy computers in unmanned vehicles will become a thing of the past, and the cabin will finally stop being hot from heating video cards,” Pavel Karasev shares his opinion.
How are the competitions going?
Formula Student is an international student engineering competition. Their goal is to form engineering thinking among students and educate specialists who are ready to solve the combat creative tasks of the auto industry.
Races are held in two stages: static and dynamic. During the first, the design of the car, its cost and business plan are evaluated, technical specifications are inspected.
Among the dynamic tests, the most spectacular is the Track Drive. Race cars move in turns and overcome 10 laps with a total length of 5 km at maximum speed without the help of pilots and remote control. The winner is the one who scores the most points in all disciplines.
Along with design and technological tasks, team members have to solve financial and personnel issues: assemble a team, organize work, attract sponsors, go through a tough selection to participate in different stages.
Competitions are popular among European universities, as they allow you to apply the acquired knowledge in practice. For example, almost every German university that trains specialists for the automotive industry has its own team to participate in such races.
Along with student Formula Student competitions, there are others in the world, for example, Roborace unmanned auto racing. And the movement of unmanned vehicles and the first serious developments in this area began with the DARPA Grand Challenge and DARPA Urban Challenge, which were held in 2005 and 2007.
What’s next?
The terms of Formula Student Driverless mean that the same drone can race for a year from the date of entry in the first races. The Bauman Racing Team expects to take part at least in Hungary in August 2021, as well as in our country in September, and possibly in other European stages this and next year.
It is very difficult to get into international competitions: there is a strict selection, and the number of participants is limited. Each trip to European races costs a team of 30 about ₽1 million – not counting the year of preparation. Sponsors provide money for participation.
After the first completed stage, based on the knowledge and experience gained, the team begins to develop a new drone.
Why do we need unmanned racing?
Racing unmanned cars, on the one hand, are safer, since no one dies or gets injured in them, on the other hand, they are devoid of the entertainment and adrenaline that are present in ordinary ones.
If we talk about Formula Student race cars, then their goal is to train new personnel for the auto industry using the example of a real drone project. Having built such a car, each student goes all the way from idea to testing and implementation.
As a result of the competition, members of such teams often receive lucrative contracts from the largest automakers that develop unmanned vehicles. Representatives of Audi, Volkswagen, Mercedes specially visit the European stages of racing to select talented personnel.