A few years ago, we were told that self-driving cars were just around the corner and would soon be pulling up at the kerbside. Then, when the fiendish complexities of developing and deploying the technology became clearer, we discovered they were stuck in heavy traffic many blocks away. But it would be wrong to conclude that fully autonomous vehicles will never arrive — cancelling on us like some frustrated Uber driver. While the challenges remain enormous, steady progress is being made.
This week, for example, the Chinese technology company Baidu announced it had won approval to operate fully driverless robotaxis in Chongqing and Wuhan on a commercial basis. It will test five vehicles in each city in controlled areas during restricted times. Baidu’s pilot robotaxis, with a human safety operator in the passenger seat, have been running around some of China’s biggest cities for many months. To date, they have amassed a total of 32mn kilometres of real world driving data to feed Baidu’s ever-improving algorithms.
Last month, Baidu unveiled its sixth-generation Apollo RT6 robotaxi with 38 sensors, including eight Lidar systems and 12 cameras. It aims to deploy up to 100,000 of these cars from next year. The company said it had slashed the production cost to $37,000, pushing the technology towards commercial viability. “We are moving towards a future where taking a robotaxi will be half the cost of taking a taxi today,” said Robin Li, Baidu’s co-founder.
In the US, Waymo, the self-driving company owned by Alphabet, has been testing autonomous cars in Phoenix, Arizona, since 2020. The company recently announced that it had learnt enough to remove human safety operators from its cars and extend its service downtown. In June, Cruise, General Motors’ autonomous car unit, became the first company in the US to offer commercial robotaxi rides in San Francisco. But it suffered an embarrassing setback when more than half a dozen immobile Cruise robotaxis blocked an interchange for a couple of hours before being manually removed.
Worryingly, the US National Highway Traffic Safety Administration this year opened an investigation into the Autopilot function installed on an estimated 830,000 Tesla cars following a series of crashes, one fatal. Investigators are focusing on whether Tesla’s driver assistance systems undermined “the effectiveness of the driver’s supervision.”
In the US, where suspicion of government regulation runs deep, there is a belief that the increasingly sophisticated “hive mind” software that runs fleets of autonomous cars will eventually overcome all obstacles — even if investors are impatient about how long it is taking. Unsurprisingly, China is pursuing a far more statist approach. Its self-driving car companies, including Baidu, are benefiting from the Chinese Communist party’s 14th five-year plan, which includes the promotion of autonomous driving as a national priority.
Wuhan has been designated as a technological development zone with a remit to accelerate the deployment of driverless cars. The city has opened 321km of “smart roads” for testing, one-third of which have full 5G coverage. Similarly in Chongqing, supporting digital infrastructure has been built on more than 30 test roads to facilitate real-time, vehicle-to-road communication.
Some industry leaders argue that the fastest progress is likely to be made in “operational design domains” — controlled environments away from ordinary traffic. For years, autonomous trucks have been operating in opencast mines and are increasingly being used in business parks and airports. These off-road markets are great places to deploy at pace and scale, argues Paul Newman, founder and chief technology officer of Oxbotica, the Oxford-based software company.
Although he is “pathologically positive” about the industry’s technological progress, Newman says autonomous car companies still confront the dual challenges of reassurance and insurance. Oxbotica’s MetaDriver software, which runs synthetic simulations of the hazardous surprises that can ambush drivers drawn from real-world data, can help reassure users and operators that the technology is safe. But insurers want additional clarity about operational responsibility and legal liability — as outlined in a recent report from the UK’s Law Commission. “We need to move towards systems that put architectural design above algorithmic obsession,” Newman says. “We have to view regulators as co-founders.”
Few quick-witted entrepreneurs relish intense talks with slow-moving bureaucrats. But the lack of public acceptability and legal certainty may yet prove the industry’s biggest roadblock. When it comes to autonomous cars, innovators and regulators have to work as frenemies.