What if there was a way to deploy autonomous vehicles prior to technological perfection? China may have found a way: build environments where AVs can operate separate from everything else. In this case, the key to success is infrastructure rather than technology.
On the one hand, AVs are not so complicated; OEMs simply bring together various hardware, software, mechanical and other technologies and make them work together. In the words of one industry executive, “It’s relatively easy to manually convert today’s automobile into tomorrow’s AV.” Indeed, universities regularly host competitions, pitting teams against each other to do just this. “The problem,” my interlocutor continues, “is how AVs behave when they are surrounded by less intelligent vehicles as well pedestrians, cyclists and so on.”
The Society of Automotive Engineers lists six levels of driving automation: Level 0 through to Level 5. For Level 0 vehicles, a human must constantly supervise and control the vehicle at all times and in all conditions; any technologies present are purely supportive and only provide warnings or momentary support (e.g. blind spot warnings). At the other end of the spectrum, for Level 5 vehicles, a human does not need to do anything; the vehicle can drive itself under all conditions and in any circumstance.
Today, the industry is focused on Level 4 capabilities; vehicles that can drive themselves but will instruct the human to take over under some circumstances. German automaker Audi’s new Q7, for example, is a Level 4 AV; while in China, software companies such as WeRide.ai and Pony.ai are focused on producing AI-driven Level 4 software technologies.
Transitioning to Level 4 is very difficult because of the fundamental shift in human-computer interaction it necessitates. There are two issues here. First, requiring a human to take control of a highly automated system is problematic. Initial findings from the terrible Boeing 737 Max crashes point to the cognitive limitations of humans trying to respond to automated systems that behave in unexpected ways. When a Level 4 AV asks the passenger to become the driver, there may only be a five-second window; not enough time for most humans to respond effectively.
Second, the problem of mixed equipage. This refers to environmental factors and the very likely scenario of Level 4 and 5 AVs operating in public with other (non-intelligent) vehicles as well as pedestrians. Japanese automaker Nissan takes this topic so seriously they set up a Human Understanding and Design group—anthropologists are among its staff—to answer the following question: exactly what constitutes “socially acceptable” behavior for an AV? Consider a crosswalk: even with clear rules of engagement, very often the driver and pedestrian will make eye contact to decide who proceeds first using subtle facial and hand gestures that AVs cannot currently recognize, let alone interpret.
China has a solution to these conundrums: deploy AVs in controlled environments while AI and software technologies develop. This extends beyond university campus buses and seaport facility trucks. While other nations must retro-fit their road network to accommodate AVs, China is still building cities and roads; it can therefore design them AV-ready from day one. The government of Zhejiang province, for example, built the country’s first “super highway” for AVs, connecting Hangzhou and Ningbo cities. This is just the beginning; China has identified more than 200 pilot cities as part of its Smart City agenda. Moreover, some industry executives are expecting Chinese authorities to implement strict regulations rendering some road lanes—even entire highways—as “AV only.”
These regulations may be necessary given the government’s aggressive targets: 50 percent of new cars must have driver assistance and partial or conditional autonomous driving features by 2020; 10 percent of all new cars must have fully autonomous capabilities, also by 2020; and Level 4 AVs need to be available for purchase as early as 2025. As socially acceptable behavior still has a long way to go for AVs, giving them their own roads seems to be the best solution for now.
Concurrent with technology developments, regulations and standards will also help the transition to full autonomy. But emerging AV standards in China are cause for concern. One standard, for example, requires autonomous buses to automatically break at speeds of 60 km/h when a pedestrian walks in front of the bus (within its 45-degree peripheral vision). This use-case, it was pointed out to me, is utterly useless on AV-only lanes and highways (where pedestrians are unlikely; but then again, this is China) as well as in mixed equipage environments, where speeds are slower and a wider peripheral vision closer to 180 degrees—or even 270 degrees—is necessary.
Nonetheless, while China continues to develop its AV behavior capabilities and experiment with standards, its infrastructure advantage may give it an edge in developing and deploying AVs faster than other nations.