Unmanned VS Manned

A pilotless vehicle either works on its own or is controlled remotely by a human driver. The two main components of such a device are the decision-making block and the machine vision system. The latter includes lidars, cameras, and other sensors.

As for the automation degree, there are several classes of driverless transport; for example, railroad transport has four classes. The first is non-automated, the second is semi-automated, when the driver doesn’t control the train’s movement, but controls the boarding and alighting of passengers, and takes full control in case of an emergency. The third class is when the train’s movement is fully automated, but the driver is still present, and the fourth means no driver on the train.

Driverless systems outdo a human driver in several important criteria. For example, the sight range of a regular human driver at daytime is about 1 km, and at night, it’s limited by how far the headlights reach. Machine vision systems have infrared cameras. Another criteria is reaction speed. For a human, it’s about 1.5-2 seconds and depends on the time of the day, as well as their physical and emotional condition. The reaction time of a driverless vehicle is always about 1 second – from detection to turning on the brakes. With every update of software, this time will continue to diminish.

Railroad transportation is part of the critical infrastructure: if a train stops on the tracks, this paralyzes the traffic and results in severe economic and political damage to a city and the country in general. At the same time, researchers state that an automated train solves the rail traffic problem. For example, the current interval between trains in Moscow Central Circle is about four minutes. The introduction of driverless trains can reduce it to three.

Safe infrastructure VS Safe transport

ITMO’s staff and students are in the process of developing validation and verification methods that will prove that the system meets the specifications and is less harmful to people and the environment than transportation with a human driver. The project’s task is to conceive and trace all possible negative scenarios that the system’s operation can result in, and find solutions.

The researchers pay special attention to the sensor module and the neural network. The latter is sometimes referred to as “the black box”: the AI creates algorithms by itself while it learns, and scientists have to make sure that every string of its code will lead to a safe outcome, that the driverless vehicle will not create dangerous scenarios.

There already is driverless transport in the USA, UAE and Japan. But in those cases, there are special safe infrastructures developed for it. For example, separate lines going two ways and fixed at a specific height. The goal of ITMO’s staff and students is to create a train that can be introduced in any infrastructure with no need to rebuild cities or construct special railways. It should be possible to use such transportation in any territory, climate, and for any purpose.

What’s more, the methods, models, and test systems that are being developed by the researchers can be applied to various vehicles. For one thing, the computer vision system of a car has very little difference with the one used in trains.

“For now, we can only speak about applying those systems where there is 100% connection quality and coverage, as well as a controlled maintenance area. Within Moscow Central Circle, we can quickly get to any point of the line: our team will check what’s wrong with the train and get it moving again. The same won’t work in case of intercity traffic, as the risks would be colossal,” comments Ilya Popov, the project’s head.

Passenger-less train: GoA3+

As of now, there’s already a train fitted with GoA3+ level automation systems, Lastochka №136, that’s used in Moscow Central Circle. It’s fully automated, but there is still a driver in the cab to neutralize the risks in case of an emergency. There are no passengers on this train yet, it’s used for gathering statistical data that’s necessary for certification and further research.

During these tests, researchers study how particular systems work, their malfunctions, the train’s behavior when it interacts with the environment, and analyze the risks. The scientists also use the data to improve various properties in order to increase the train’s automation level to the highest.

For the time being, the driverless train is configured in such a way that it’s highly sensible to all factors. And if the system doesn’t know how to react to something, it’s likely to signal and brake. In this case, the person in the cab will take control and decide whether to stop or move on. This data is used for teaching the neural network.

GoA3+ wasn’t used on a regular subway line due to a logistics issue: the driverless equipment is in the head of the train, so it would’ve been necessary to have a person drive it the other way. There are no such problems with a circle line.

Driverless supremacy: GoA4

GoA4 is the next level of automation, when there’s no human driver in the train. In this case, the train is controlled remotely, and only in case of an emergency on the system’s request. At the control center, there is only the operator with the same controls as those in a train, it’s just they will have displays rather than a windshield in front of them. When they get a signal from the train, it would be up to them to decide whether to take control or not.

At the Ural Locomotives plant, they are already manufacturing the first Lastochka train that’s completely automated. Its driver’s cab will be altered to store computers. This train will be a prototype, and will be mass-produced if certified. If everything goes well, GoA3+ trains will be introduced in 2022, and the transition to GoA4 will become possible in two or three years.

“After a completely safe train will start running, we’ll face another difficulty: some passengers will fear and distrust the new technology. Sceptics will be against riding driverless trains. But there once was a time when people were against even simple machinery, so I guess this is inevitable,” says Ilya Popov.

Tricking the machine: how to test a train at a university

To prove that a driverless train’s operation is safe, it’s necessary to test the machine vision system. For that purpose, ITMO and JSC NIIAS are collaborating on creating an experimental laboratory. The system will contain a multitude of cameras, lidars and other sensors, fixed on a pole and located on a vibration stand that simulates mechanical effects. On the other side, there will be display targets that simulate real objects. Various generators will help simulate changes in climate conditions.

At such a laboratory, it will be possible to measure things like how thick mist affects the sensors, i.e. how will it hinder their range and ability to detect objects. And a lighting control system will help model daytime and nighttime conditions or show how foggy the images will be at sunrise or sunset.

“As of today, it’s stated in the Vienna Convention that every moving vehicle or combination of vehicles must have a driver. There are plans to make changes so that it would be possible to use driverless cars without a driver on public roads. In collaboration with the Russian Railways and JSC NIIAS, we are currently working on the legal regulation of driverless vehicles in railway logistics. By the way, we already make use of specific aspects of law in what has to do with driverless cars,” says Ilya Popov.