Tesla cars frequently have accidents, how safe is autonomous driving?

　　Recently, the rights protection at the Shanghai Auto Show and the death of one person after a Tesla crashed into a concrete wall in Guangzhou have pushed Tesla to the forefront of public opinion. On the 27th, the topic of "A Tesla crashed into a steamed bun shop in Jiangsu" became a hot search on Weibo. Coincidentally, a Tesla car crash also occurred in the United States recently, resulting in the tragedy of two deaths.

　　Regardless of the cause, the incidents have raised questions and debates about today’s self-driving technology: How safe is it, how much attention does it require from drivers, and will it still be possible to buy self-driving cars in the future?

　　An article recently published in Australia's Dialogue magazine believes that although the technology required to achieve a higher level of automation is developing rapidly, it is still a huge challenge to produce a car that can safely and legally travel the entire journey without the driver's attention. Before they can safely enter the market, they must overcome three key obstacles: technology, regulations and public acceptance.

　　How to understand "autonomous driving" technology

　　First, it is important to understand what "autonomous driving" technology is. There are six levels of autonomous vehicle technology, ranging from Level 0 "no automation", which is a traditional vehicle without autonomous driving capabilities, to Level 5 "full automation", which is a vehicle that can independently do everything a human driver can do.

　　Most self-driving car operations currently on the market require human intervention, such as Level 1 vehicles with “driver assistance,” which keeps the vehicle in its lane or controls its speed, or Level 2 vehicles with “partial automation,” where the driver must be on hand to steer and control speed at all times.

　　Level 3 vehicles have more autonomy, with the car making some decisions on its own, but the driver must still remain alert and take control if the system fails to drive.

　　Level 4 and 5 vehicles have a higher level of automation, and a human driver is not necessarily involved in the driving task. Vehicles at these two levels are able to steer, brake, accelerate, monitor the vehicle and the road, and respond to events, determine when to change lanes and make turns.

　　But Level 4 vehicles are limited in where and when they can drive. Level 5 represents truly autonomous vehicles that can drive anywhere, at any time, similar to human driving. However, the transition from Level 4 to Level 5 is more difficult than the transition between other levels and may take years to achieve.

　　Machines should "learn" a lot about actual driving scenarios

　　Self-driving software is a key feature that distinguishes highly automated vehicles from other vehicles. The software is based on machine learning algorithms and deep learning neural networks, which include millions of virtual neurons that simulate the human brain.

　　Neural networks need to be trained to learn to recognize and classify objects using millions of examples of videos and images from real driving conditions. The more diverse and representative the data, the better the neural network can recognize and respond to different situations. Training a neural network is a bit like holding a child's hand while crossing the street, teaching them to learn patiently through experience and repeated training.

　　Although these algorithms can detect and classify objects very accurately, neural networks still cannot mimic the complexity of real driving. Self-driving cars must not only detect and recognize people and other objects, but also interact with, understand and react to the behavior of these objects. They also need to know what to do in unfamiliar road conditions. Without a large number of examples for all possible driving scenarios, deep learning and training for unexpected events becomes relatively more difficult.

　　Vehicles should be rigorously assessed before they go on the road

　　Policymakers and regulators around the world are struggling to keep up with the development of autonomous vehicle technology. Today, the industry is still largely self-regulating, especially when it comes to determining whether the technology is safe enough and suitable for use on open roads. The article in The Conversation magazine stated that regulators have largely failed to provide standards in these areas.

　　It is necessary to test the performance of autonomous driving software under real-world conditions, starting with comprehensive safety testing and evaluation. Regulators should develop a set of standard testing protocols to allow companies to benchmark their algorithms against standard data sets before their vehicles are allowed on the road.

　　In Australia, current laws do not support the safe commercial deployment and operation of autonomous vehicles. The National Transport Commission is leading a national reform effort to support innovation and safety in autonomous technology so Australians can enjoy the benefits of this technology.

　　The article argues that a progressive certification approach is needed when it comes to regulations for autonomous driving technology. This approach should require that autonomous driving systems be evaluated first in simulations and then in controlled real-world environments. Only when vehicles pass specific benchmark tests can regulators allow them to drive on open roads.

　　Public acceptance is key to trust in technology

　　The public also needs to be involved in decisions about the deployment and adoption of autonomous vehicles. If autonomous technology is not regulated to ensure public safety, there is a real risk that public trust will be undermined. Lack of trust will not only affect those who want to buy autonomous vehicles, but also those who share the same roads with these people.

　　Finally, recent incidents should serve as a catalyst for regulators and the industry to establish a strong safety culture to guide innovation in autonomous driving technology. Otherwise, autonomous vehicles may have a bumpy road ahead and will not be able to go far.