Smarter than humans – deciphering AI in driverless technology

Driverless cars are a hot technology at the moment, with both concept cars and real driverless cars in production appearing, and the development of AI technology is pushing the envelope, making driverless cars a qualitative leap forward. This article explains how AI plays a role in driverless cars, and whether it has really surpassed humans in terms of driving skills.

The history of driverless cars

Driverless technology has actually been around for a long time. As early as August 1925, a “seemingly” driverless car appeared on the streets of the United States for the first time, with a U.S. Army electrical engineer named Francis P. Houdina controlling a car by transmitting radio waves. driverless concept car (Figure 1).


Figure 1 The first driverless car

With the development of science and technology, especially the development of AI technology, more and more technology giants are involved in the development of driverless technology. 2015, Google’s first prototype car was officially unveiled and officially tested on the road, during which the car could automatically avoid pedestrians, intelligently identify traffic lights, automatically meet with other oncoming vehicles, etc., just like a skilled old driver in In 2018, driverless technology has further developed, and the integration of AI technology has added wings to the driverless car, further improving the driving level.


Figure 2 Google’s first driverless prototype car

AI in driverless cars

We introduced above that driverless cars can automatically avoid, rendezvous and recognize signals, etc. So what exactly is the role of AI in the whole driving journey?

We all know that driverless cars must be accurately positioned and controlled in order to drive freely on the road. Conventional positioning uses the data acquired by the optical radar and camera on the driverless car, and then realizes the positioning of the car on the map through certain algorithms. However, since driverless cars need to drive through complex geographic environments and ever-changing streets, this requires a very good perception and decision making capability of the driving system, which obviously has a very high uncertainty in itself (Figure 3).


Figure 3 Optical radar and cameras on driverless cars

In order to achieve more accurate positioning and more intelligent decision making, scientists use AI’s deep learning to solve this problem. Instead of pre-designing the algorithm first like traditional deep learning models, they prepare a large number of examples for the driverless system. On the one hand, the driving system acquires more complex road driving techniques and decision-making capabilities by learning a large number of examples, and on the other hand, it improves itself by learning from the examples and constantly correcting itself through the AI’s autonomous learning capability. The whole process is similar to experienced and diligent drivers who can improve their driving skills and their ability to deal with various road emergencies by communicating with and learning from other colleagues (Figure 4).


Figure 4 AI deep learning illustration

At the same time, the researchers also used “multi-task deep learning” to train the driving system in depth. In this training, the researchers let the driving system recognize lane markings, cars and pedestrians at the same time, and further improved the driverless system’s recognition and judgment ability by using AI’s CNN (Convolutional Neural Network) technology.

With the deep learning mentioned above, driverless technology now does not rely on a predetermined map, but uses the map as one of the data streams, and combines the data obtained from sensors to help the system make decisions. For example, the neural network system can use the map information to know the information of the next pedestrian crossing in advance, and then use the information of pedestrians crossing the crosswalk captured by the camera in real time, so that the driving system can make better decisions, such as stopping to yield to pedestrians or crossing the crosswalk normally. With the “multi-tasking deep learning” AI neural network, there is no need to extract lane information, traffic signs and vehicle-pedestrian signs from the original pixel map, it is all left to the neural network to automatically identify them, and finally just output simple instructions such as braking, steering and yielding for the driverless system to execute (Figure 5).


Figure 5 Illustration of the application of neural networks in driverless

Of course there are many applications of AI in driverless technology. For example, AI deep learning allows the driving system to learn many human driving styles and skills, thus making passengers feel like an experienced driver driving and thus improving the ride experience. In addition similar global navigation path planning aspects, local environment map three-dimensional construction, intelligent deployment of vehicles, etc.

Driverless changes your life and mine

Through the above introduction, we know that driverless cars using AI technology can not only improve the safety of driverless, but also cope with more complex road conditions and have better decision-making ability than humans. This not only adds a lot of selling points for manufacturers, but also makes a lot of sense for ordinary users.

Statistics show that 1.2 million people die in car or traffic-related accidents worldwide each year, 93% of which are caused by human error. Driverless cars make full use of AI technology to reduce exactly those traffic accidents caused by human errors caused by carelessness, and thus also reduce the number of casualties. As an example, the Google Waymo driverless car, which has been tested for eight years, has driven more than 3 million kilometers during that time and has only had a dozen minor accidents (Figure 6). Although a Google Waymo driverless car was involved in a crash in the United States in May 2018, it is worth noting, however, that the Waymo driverless car was not the at-fault party, but rather the party to which the accident was caused.


Figure 6 Google Waymo driverless car all-round monitoring to enhance safety

The popularity of driverless technology will undoubtedly bring more convenience for ordinary people to travel. When the whole driverless technology is maturely applied to the public transportation system, all the traffic will be unified in one system, so that no matter where you are and when you take a taxi, driverless vehicles will provide you with the best service according to automatic scheduling.

Of course, there are many applications of driverless technology, such as we drive to the destination without having to consider the difficult problem of parking, it will automatically help us to find a parking space and so on.

Leave a Comment