WATCH | Inside a robotaxi: a Tokyo test ride

Self-driving vehicles don’t suffer from impatience or road rage, and their eventual mass rollout is expected to significantly improve road safety, given that the vast majority of collisions are caused by human error.

Robotaxi fleets are operating commercially in several cities across the US, China, the United Arab Emirates and Croatia. Some have up to Level 4 autonomy, which means they are able to drive themselves in geo-fenced areas that have been mapped in high definition. Some of these vehicles have no human in the car but are monitored remotely by human operators who provide guidance if the AI encounters a scenario it can’t handle.

Full Level 5 — the holy grail where a car can be used anywhere without a steering wheel — has proven harder and more expensive than initially expected.

A “system failure” caused a robotaxi outage involving many vehicles operated by Baidu’s Apollo Go in central China’s Wuhan, local police said on Wednesday, reigniting safety concerns over the fast-growing service. Police received reports late on Tuesday that many Apollo Go cars had stopped in the middle of roads in Wuhan and were unable to move, according to an official statement. Passengers were able to exit the vehicles safely and there were no injuries, police said. The cause of the incident is under investigation. At least 100 Apollo Go vehicles were affected, a traffic police officer said in a video published by Shanghai-based news outlet The Paper. The officer said while the car doors could be opened, some passengers were hesitant to get out because of heavy traffic and called police for assistance. Local media reported some passengers were trapped inside vehicles for nearly two hours. Baidu did not immediately respond to a request for comment. The accident sparked renewed discussions on Chinese social media about robotaxi safety and readiness. An Apollo Go robotaxi carrying a passenger fell into a construction pit in Chongqing last year August, and in May a car operated by Pony.ai caught fire on a road in Beijing. No injuries were reported in either incident. A widespread power outage in San Francisco at the end of last year also caused Waymo robotaxis to stall and snarl traffic. Baidu is one of China’s largest operators of autonomous driving fleets, alongside Pony.ai and WeRide. The companies have rolled out commercial robotaxi services across major Chinese cities and have expanded operations into overseas markets, including the Middle East.

There have been a number of well-publicised accidents caused by self-driving cars, and some carmakers are also questioning whether there will be enough customer demand for autonomous vehicles to justify the investment. Ford and GM, for instance, have abandoned their in-house robo-car projects, but others are betting on advances in artificial intelligence to give the technology a game-changing boost.

Nissan, Uber and Wayve are using AI to develop robotaxis together with the intention of rolling out a pilot programme in Tokyo by late 2026. Nissan has also begun testing a driver-assistance system using the AI technology of British startup Wayve, ahead of a planned launch in select vehicles in Japan in 2027.

Business Day was invited to experience a demonstration of Nissan’s AI-enhanced ProPILOT system in Tokyo this week, riding in a self-driving prototype based on the Nissan Ariya. The car uses 11 cameras, five radar sensors and a next-generation LiDAR sensor to keep digital eyes on the surroundings, detecting objects at a distance in day or night-time driving, in slow or high-speed conditions.

With a safety driver seated behind the wheel as backup, the Ariya drove autonomously on a 30-minute route through the capital’s busy streets. The car was designed to navigate intersections and complex urban environments, and to understand how driving scenes evolve rather than just reacting to single objects. It can adapt to real-world scenarios and make decisions in complex situations.

The short history of the self-driving car industry has been littered with expensive failures and endless delays, but tech suppliers, chipmakers including Nvidia and some carmakers are betting on AI and a web of partnerships to spark new progress. Many interested carmakers, however, still have major questions. Apart from concerns about high costs and scalability, they want to know if there is enough customer demand to make money out of an expensive wager. Vehicles that drive themselves would change the transportation landscape, but making such a technology safe for public roads has been harder and much more expensive than expected. While a few companies such as Alphabet’s Waymo and Tesla have decided to do it themselves, veterans such as General Motors and Ford have abandoned their in-house efforts for fully autonomous vehicles. At the CES show in Las Vegas last week, AWS and German supplier Aumovio announced a deal to support the commercial rollout of self-driving vehicles, while autonomous truck firm Kodiak and Bosch said they have partnered to ramp up manufacturing of autonomous trucking hardware and sensors. AI chip company Nvidia rolled out its next-generation platform, which will be used in a robotaxi alliance announced by Lucid Group, Nuro and Uber. Powered by Nvidia’s chips, Mercedes-Benz said this week it will launch a new advanced driver-assistance system in the US later this year that lets its vehicles operate autonomously on city streets under driver supervision. Jochen Hanebeck, CEO of German chipmaker Infineon, cautioned against the “market fantasy” that somehow fully self-driving cars could become commonplace within a few years The propulsive force behind self-driving technology — artificial intelligence — is also coming into its own as a development tool, offering hopes of mitigating high costs. AI and generative AI are acting as a “big accelerant” for the industry “because it actually allows ... a significant amount of development and validation with significantly fewer resources”, said Ozgur Tohumcu, GM for automotive and manufacturing at Amazon’s cloud unit Amazon Web Services. Western carmakers are also under pressure to keep up with China’s push to lead the development and adoption of autonomous driving. Last month, the Chinese government approved two cars with Level 3 autonomous capabilities, which allows hands-off driving. The car industry has defined five levels of autonomous driving, from cruise control at Level 1 to fully self-driving, without a human minder needed, at Level 5. Still, Jochen Hanebeck, CEO of German chipmaker Infineon, cautioned against the “market fantasy” that somehow fully self-driving cars could become commonplace within a few years. Rather than risk fresh investments in fully self-driving cars, major carmakers want revenue-generating driver assistance technology, known as Level 2, that is already available but requires drivers to pay constant attention, he said. “I don’t see, really now, a tsunami flowing towards Level 5,” Hanebeck said. In recent months there has been a flurry of small robotaxi deployments announced in China, the US, Europe and the Middle East, but Jeremy McClain, head of system and software at Aumovio’s autonomous mobility unit, said expanding the areas they cover requires more data, fleets and logistics, “which is costly and expensive”. Tesla CEO Elon Musk promised in 2019 that a year later the electric vehicle maker would have a million self-driving cars on the road. But Tesla launched a small robotaxi trial service only last year, six years after Musk’s bold prediction The self-driving car industry is long on hype. Tesla CEO Elon Musk promised in 2019 that a year later the electric vehicle maker would have a million self-driving cars on the road. But Tesla launched a small robotaxi trial service only last year, six years after Musk’s bold prediction. The problem was that cars face billions of potential unexpected incidents, or “edge cases”, that can easily fool self-driving vehicles. One example often touted by experts is that if a human driver sees a ball rolling into the street, they automatically slow because it might be pursued by a child, but a self-driving car will react only when it sees the child. After the first self-driving bubble burst, major carmakers, including Ford and GM, abandoned money-losing autonomous vehicle units. The demise of GM’s Cruise was accelerated by an incident in which it struck and dragged a pedestrian 6m. But Ali Kani, GM of the automotive team at Nvidia, said AI has enabled advances to address key weaknesses in self-driving technology. “There’s some foundational pieces of technology that make us feel like we’re there,” Kani said. Morgan Stanley analysts in a note on CES said that while Nvidia’s new Alpamayo platform for self-driving would give legacy carmakers a leg up and help them pressure Tesla, the EV maker was years ahead. That said, many in the industry see Nvidia, whose platform is open-source, as a convenient place for Tesla rivals to gather. “In one way, you could almost see Apple and Android playing out,” said Russell Ong, former product lead at self-driving vehicle maker Zoox, referring to Tesla’s proprietary system vs Nvidia’s decision to release Alpamayo as an open-source model. Reuters

The Nissan drove with a consistency and restraint expected of a well-trained human driver, showing consideration for other road users. Impatience wasn’t in its repertoire, and it didn’t try to steal gaps with a me-first attitude. It maintained safe following distances, changed lanes without cutting in front of fast-moving traffic, and prioritised pedestrians in its cruise through the city.

The car also coped with unexpected situations. When encountering a stationary vehicle partially obstructing its lane, it correctly identified the obstacle and overtook, rather than assuming it was part of slow-moving traffic.

Thanks to conflict in the Middle East and high oil prices ($96.12, or R1,572.73 a barrel at the time of writing), South African motorists should brace for even higher fuel costs in May, with the latest data from the Central Energy Fund (CEF) showing growing under-recoveries across all fuel grades. The CEF’s latest daily snapshot, published on April 14, points to petrol price increases of about R2.62/l for 93-octane and R2.99/l for 95-octane. Diesel users face a far steeper rise, with 0.05% sulphur diesel expected to increase by R9.05/l and 0.005% sulphur diesel by R9.07/l. Currently 93-octane petrol, available only inland, is priced at R23.25/l, while 95-octane petrol costs R23.36/l inland and R22.53/l at the coast. Wholesale diesel prices stand at R25.90/l for inland 0.05% sulphur diesel and R26.11/l for 0.005%, compared with coastal prices of R25.07/l and R25.35/l respectively. Adding further pressure is the expected end to the government’s temporary R3/l reduction of the general fuel levy across all grades of petrol and diesel. This would compound the projected increases, potentially pushing petrol prices towards R30/l and diesel closer to R40/l. As always, the projections are subject to change before month-end, with the final fuel price adjustments for May set to take effect on May 6.

The technology is not infallible and, on one occasion, the car activated its indicator a street too early — an error that could confuse other road users.

Overall, the ProPILOT system delivered an impressive performance on Tokyo’s well-maintained roads and among its generally disciplined drivers. It remains to be seen how AI-enhanced cars might cope outside of first-world driving conditions, for instance in South Africa which presents more challenges with its potholed roads, erratic driving behaviour and malfunctioning traffic lights.

Business Day