From driverless cars to smart manufacturing at the factory, AI is transforming the automotive industry globally. Two feature articles look at AI and auto industry. The first of the two articles look at the self-driving technology and AI, to be followed by the second article on the AI and car manufacturing.

Development and commercialization of self-driving cars in Japan and globally

The Society of Automotive Engineers (SAE) defines 6 levels of driving automation ranging from Level 0 to Level 5. The majority of vehicles on the road today are defined as Level 0, in which all aspects of driving are fully human and manually controlled. Level 5 vehicles do not require human attention as manual driving is completely eliminated at this level.

In Japan, some manufacturers have developed Level 2 cars. The Advanced Driver Assist System such as Nissan’s pro pilot allows the vehicle to automate certain parts of the driving experience, but the level of automation falls short of self-driving because a human driver still sits in the driver’s seat and can take control of the car at any time. To accelerate the development and commercialization of higher level of self-driving cars, Japan passed legislation last year to allow Level-3 self-driving cars on the road. The new laws will go into effect this spring, and it is reported that Honda set to launch a partial self-driving car this summer. Honda’s Legend sedan will be equipped with a Level-3 autonomy system, allowing the vehicle to pilot itself for extended periods. Also, Toyota will supply “Tokyo 2020 Version” e-Palette vehicles to support athlete mobility at the Olympic and Paralympic Games Tokyo 2020, providing automated, loop-line transportation in the Olympic and Paralympic villages for athletes and related staff. This specially-designed vehicle is equipped with control hardware, software, and advanced sensors such as cameras and LiDAR and will provide will realize low-speed automated driving at Level-4.

In the U.S., Waymo, the self-driving vehicle venture of Google parent company Alphabet, launched a commercial autonomous taxi service Waymo One. The service is available for Phoenix residents who were part of its early rider program. The users can use the app to inform the pick-up point and the destination, and this Level-4 self-driving taxi will do the rest. Initially, a safety driver was sitting behind the steering wheel just in case, but in November 2019, Waymo One launched that fully-driverless service in Arizona.

Volkswagen Group announced set up a subsidiary called Volkswagen Autonomy (VWAT) in 2019. In addition to the offices in Germany which focus on research and development of Level-4 autonomous cars, it also plans to establish companies in Silicon Valley and China in 2021 with the goal of making autonomous driving market-ready. It has been reported that Volkswagen Group plans to bring robot taxis and cargo vans to three continents by 2025.

In an effort to keep pace with the rivals and lead a global race to develop autonomous vehicles, China has laid out national guidelines for testing self-driving cars on the road. Chinese cities including Beijing and Shanghai have also announced local guidelines for self-driving tests, and Baidu already has approval to test Level-4 self-driving vehicles on city streets. In September 2019, Wuhan, capital of China’s Hubei Province, issued commercial licenses allowing Baidu, Haylion Technologies and DeepBlue Technology to conduct trial operation of passenger-carrying self-driving cars. It is reported that Baidu got a license for five of its self-drive cars, while Shenzhen Haylion Technologies Co. and DeepBlue Technology will trial a bus each.

AI behind the high-level autonomous driving

A range of advanced technologies ranging from location detection and identification, AI, big data to communication, are used to realize advanced self-driving. To navigate through the traffic and handle complex situations, the self-driving car uses positional information from the GPS and inertial navigation system to localize itself and sensor data to refine its position estimate as well as to build a three-dimensional image of its environment.

GPS can have errors up to a few meters, but with High Definition (HD) maps (with accuracy of object locations up to 10 cm), the localization software can compare features of surroundings with the map and figure out where exactly it is located. Cameras and LiDAR act as an eye of the self-driving vehicles, providing them a 360-degree view of the surrounding helping and helping the vehicles travel smoothly and avoid collisions by detecting the obstructions ahead.

There must be a navigating system that is capable of incorporating all the inputs and take necessary actions to navigate through constantly changing surrounding environment. Waymo’s AI algorithms, for example, are fed with real-time data from sensors, GPS, radar, LiDAR, cameras, and cloud services. These data are processed to produce control signals that are used to operate the car. In other words, powerful AI system is essential for achieving high-level self-driving.

That is why deep learning is so important. The control system of self-driving cars receives a large amount of data from sensors and cameras, and the system has to be able to detect obstacles, decide how these obstacles behave, and operate the cars to avoid the obstacles. The problem is that every obstacle is different in size, shape and speed. The system must be able to perceive different types of cars as cars, and different pedestrians as pedestrians. Deep learning, which is a type of machine learning that uses lots of layers in a neural network to analyze data at different abstractions, is the perfect tool for improving the perception and behavior of self-driving cars. Using deep learning, large neural networks are trained for tasks such as image classification, object detection, and driving lane detection. The networks are then optimized for the car’s computing unit to be able to handle the real-time speeds required for self-driving. Waymo’s engineers, for example, are using deep learning to interpret, predict, and respond to data accrued from its 10 million miles driven on public roads and 5 billion driven in simulation.

This requires high computing power, and companies such as NVIDIA are developing a computing platform that can handle Level-4 and even Level-5 self-driving. NVIDIA DRIVE AGX Orin, a highly advanced software-defined platform for autonomous vehicles and robots, introduced in December 2019 is one example. The Orin system-on-a-chip integrates NVIDIA’s next-generation GPU architecture and Arm Hercules CPU cores, as well as new deep learning and computer vision accelerators. With the capacity to deliver 200 trillion operations per second, it can be used in full self-driving Level-5 self-driving cars.

An industry-wide collaboration is also accelerating. Today, over 150 members – automakers, tier 1 suppliers, component producers, startups, academic institutions and government departments – participate in Baidu’s Project Apollo, an open source platform for self-driving that includes hardware, software and cloud data services for autonomous vehicles. Toyota joined the partnership in 2019.

Challenges ahead

From technology point of view, Level-5 self-driving cars appear to be on sight. However, obtaining society-wide consensus in terms of ethical questions associated with the self-driving cars may prove to be more difficult than technology-related challenges. Questions of what a self-driving car ought to do if it encounters a situation analogous to the ‘trolley problem’ have dominated recent discussion of the ethics of self-driving cars.

In 2018, an Uber testing vehicle hit and killed a woman – the first pedestrian fatality caused by an autonomous vehicle. There are also several litigations around fatal accidents while using Tesla’ autopilot features, which is Level-3. There is no simple answer about who is at fault when these vehicles cause deaths. It turns out that answering this question of liability is just as tricky as understanding the autonomous technology. Globally, some 1.35 million people die each year as a result of road traffic crashes, and self-driving cars are expected to reduce the numbers as most accidents are caused by human errors. However, there is no simple answer as to what level of safety is acceptable for self-driving cars. Different countries or cultures may have different answers.

The next feature looks at AI used in the design and production of cars.