Letting Go of Steering

China’s search engine giant Baidu has unveiled an autonomous vehicle with a detachable steering wheel, with a view of using the new car as a robo-taxi service in China in 2023.

The company says the cost per unit for the new model will drop to 250,000 yuan ($37,031.55) for the new model. News agency Reuters reports that it previously sat at 480,000 yuan for the previous generation. “This massive cost reduction will enable us to deploy tens of thousands of AVs across China,” Baidu’s chief executive Robin Li said at the Baidu World conference. “We are moving towards a future where taking a robo-taxi will be half the cost of taking a taxi today.”

In response the US National Highway Traffic Administration Safety (NHTSA) comments on how the progress of ‘letting go of steering’ in autonomous, automated vehicles applies in the States. Firstly, there is no vehicle currently available for sale that is fully automated or self-driving. A spokesperson from the NHTSA points out: “Every vehicle currently for sale in the United States requires the full attention of the driver at all times for safe operation. While an increasing number of vehicles now offer some automated features designed to assist the driver under specific conditions, these vehicles are not fully automated.”

US safety standards

A present, in the US, the NHTSA says vehicle manufacturers have to comply with the Federal Motor Vehicle Safety Standards, “or seek and receive appropriate exemptions, and also certify that their vehicles are free of safety defects”. This permits a limited number of automakers to test, research and pilot programs with automated driving systems on public highways. The NHTSA monitors their safety performance through its Standing General Order. The spokesperson adds that automated driving systems are also tested by the companies that develop them.

The NHTSA supports the development of automotive technologies, including driver assistance technologies and automated driving systems, as they believe they offer significant safety benefits. It adds: “In March, NHTSA issued a final ruling to help increase the safety of occupants in automated vehicles that do not have the traditional manual controls associated with a human driver. This rule clarified that NHTSA’s crashworthiness safety standards related to steering wheels and columns if-equipped standards. This means they do not require a vehicle to have a steering wheel or column.”

Complexity of autonomy

Accenture’s global automotive lead, Axel Schmidt says Level 4 autonomous driving, where the vehicle drives itself in specific set conditions without human intervention, is one of the toughest issues to solve at present: “With all sorts of sophisticated radar and sensor systems, it’s technically possible but the lack of common sense makes deep learning systems insufficiently robust.”

“They are also easy to manipulate by an adversary, for example with fake road signs. That’s why most driverless cars, buses and commercial vehicles operate in designated areas only such as certain parts of a city, certain types of streets or areas with no public traffic such as logistics centers or harbors. Many manufacturers also focus on autonomous driving in selected operational areas only such as valet parking.”

AI breakthrough limitations

He says deep learning, created by a breakthrough in artificial intelligence technology around a decade ago, led many to believe that intelligent vehicles can be trained to make the right decisions while driving. An example is Tesla’s shadow mode, which compares the vehicles decisions with those of the driver in charge of the car or against virtual testing environments. “They are two examples of how algorithms can be trained to deal with the most common traffic situations,” he explains before adding: “Yet, taking the right decisions when dealing with edge cases is still too complex.”

This complexity means that from most countries still require an emergency operator to take control of the vehicle whenever deemed necessary to prevent an accident. This is a legal requirement, which applies when a situation is too complex for the car to handle. Yet, there are some countries who’ve paved the way for Level 4 autonomous vehicle driving specific circumstances, such as a maximum operating speed of 37mph or in designated areas.

Regulating Level 3

For now, most regulators permit Level 3 autonomy on public roads, which means that a human driver retains control in critical situations. Schmidt explains: “China’s government, however, is keen to commercialize autonomous driving and is, therefore, supportive of new regulation. The country is the largest car market in the world and if autonomous driving will ever have a breakthrough, it will be here.

“Globally, however, Germany was the first country to create a regulatory framework for Level 4 autonomous driving on public roads in 2021.  In China, there is no national regulation yet but some of the country’s largest cities have introduced pilot programs. The focus is on autonomous buses and self-driving taxis, however. Others are following suit, for example Japan, which is planning to allow Level 4 autonomous cars on its roads as of April 2023. Most countries will probably have introduced a legal framework for fully driverless cars within the next five years.”

Even if a country has progressive regulation, many car manufacturers will still be careful with their roll-out of fully driverless fleets as liability issues and the risk of cyber threats remain prominent obstacles. Instead, most car manufacturers are currently focusing on Level 3 autonomous driving and collecting as much real-world data as possible through their fleets to prepare for the shift to Level 4. Progressive regulation on autonomous driving is seen as a question of competitiveness. The earlier driverless cars are allowed on public roads, the more real-world data they can collect.

Software-driven vehicles

It’s important to note that no automaker has been able to completely master autonomous driving in complex real-world conditions with absolute reliability. Schmidt believes there will always be edge cases where an algorithm is not trained to handle a situation or cannot do so on its own. Automakers are, nevertheless, going to send a large number of algorithm-equipped cars into the market as the vehicles themselves are going to become increasingly software-driven.

Real-world data is essential to making autonomous vehicles happen, requiring data to be collated from a diverse set of traffic scenarios. Schmidt comments: “The real currency is not just data, it is also the collection of various edge cases and the bigger the fleet, the better and more diverse data that can be collected. Baidu started its robo-taxi service in ten Chinese cities since 2020 and has certainly gathered a lot of data. That’s why the race for fully driverless cars favors the large manufacturers who have millions of vehicles on the roads collecting data while they drive.”

Hands-off the steering

So, when will steering wheels no longer be an essential part of the driving experience, whether for taxis or for other kinds of vehicle? Well, predictions are often unreliable. Not long ago the automotive industry thought that fleets would be fully autonomous at Level 5 by 2025 and commonplace on highways in many parts of the world. The trouble is that autonomous driving is yet to live up to its promise so Schmidt believes that their development will be more evolutionary that has been previously thought. They won’t represent a Big Bang disruption.

For now, he thinks that most automakers will invest iteratively in AI and focus on Level 3 autonomous driving while including some Level 4 features such as automated valet parking and smart summon. “We recommend that manufacturers invest in the technology early but in careful stages. Given today’s limitations, we believe that Level 3 and Level 4 vehicles will only account for a cumulated share of about 5% of the global automotive market by 2030.”