This is how Mercedes and Bosch are preparing fully autonomous driving

Fully autonomous driving is closer than it seems. Mercedes-Benz and Bosch are already working with the technology needed to reach the safety levels required for it to become a reality.

To deliver fully autonomous driving in the future, the car will have to be able to see the way a driver would. That is precisely what Mercedes-Benz and Bosch are working on: a group of sensors whose combined information allows a model of the environment to be reproduced that meets the safety requirements both manufacturers have set.

What are the keys to level 5 autonomous driving, in which the vehicle takes over driving in a fully automated way in any situation, although the driver can take control at any moment? Here are some of the most important ones.

Sensors that see everything

The first key to autonomous driving is generating that environment around the vehicle so that it can be interpreted and the car can make decisions: when to turn, brake, accelerate… To do this, it is necessary to combine the properties of different types of sensors.

• Radar sensors. They detect objects and measure their position and speed relative to the vehicle. They do this by comparing the amplitude and phase of the radar signals the sensor emits, which bounce off objects in the environment. They have a range of about 250 metres, with a 360-degree angle.
• Ultrasonic sensors. They are based on sonar technique (the one used by bats or submarines), sending ultrasonic pulses that bounce off obstacles. They serve to recognise the nearby environment, up to about 6 metres, and are used in low-speed manoeuvres, for example when parking.
• Video sensors: The Bosch stereoscopic video camera that Mercedes already uses in its vehicles provides a measuring range of more than 50 metres. It registers objects, recognises empty spaces and can even recognise partially obscured pedestrians, or interpret letters and numbers on traffic signs.
• Lidar sensor: Bosch works with this type of sensor in its autonomous car prototypes as a complement to the previous sensors. They emit infrared laser light beams, which are reflected by objects and return to the lens of the sensor, making it possible to determine in real time the distance between each point where the laser light falls and the sensor on the vehicle. The first brand to use this type of sensor in a production car has been Audi in its Audi A8.

Duplicated systems for greater safety

Another important safety-related issue is the possibility of failure. To address this, the requirements that manufacturers must set, as Daimler AG does, seem clear: having duplicated systems in all actuators that control functions such as steering, acceleration and braking.

In the case of Mercedes-Benz, this means that its autonomous cars will have not only two steering motors but also duplicated control electronics. The same goes for the brakes. Through the iBooster electromechanical brake system developed by Bosch, together with the ESP system, the ability to bring the car to a safe stop is ensured even if an individual component fails.

Artificial intelligence and neural networks

Two terms that sound like they come from a neurologist’s consultation, but to which we will have to get used. The head of image semantics at Daimler AG clarifies some concepts on this subject. For example, he points out that when people talk about the use of artificial intelligence or deep learning in cars, it does not mean that cars are going to develop a consciousness of their own.

These are not machines capable of thinking. The concept “artificial intelligence” or “machine learning” refers to very complex programming that makes it possible to use, process and interpret all the data received by the sensors we have discussed. Programs so sophisticated that they may seem intelligent. Applied to the car, it means the system is able to know what is right or wrong, what is correct or incorrect, because the programmer has set it that way, based on the evaluation of many traffic situations.

The computers of an autonomous driving system in a vehicle could learn individually by modifying their algorithms, but this is not the ultimate goal, since engineers would not be able to reproduce a possible behavioural error. The idea is to develop learning programs that can be applied to all vehicles and reproduced so that they all drive using the same logic. Programmers can make an autonomous vehicle capable of interpreting the subsequent movement a pedestrian or cyclist may make. Or whether what crosses in front of the vehicle is a ball behind which a child may follow… or a stone, or simply a crumpled ball of paper blown by the wind.

Once the goal is reached that vehicles can respond instantly in the correct way even in the most complex situations, and act accordingly, the advantage of these systems is that the machines and systems in charge of autonomous driving neither get tired nor lose concentration.