Elon Musk's Tesla is betting on “black box” artificial intelligence technology for robotics

Tesla's goal on Thursday night was to stun investors with its long-awaited “robottaxi unveiling,” a potential milestone after a decade of failed promises by Elon Musk to deliver autonomous vehicles.

The carmaker unveiled a so-called “Cybertaxi” prototype – which Musk described as capable of “fully autonomous driving under supervision” – instead of a ready-to-drive driverless taxi.

“There's no steering wheel or pedals, so I hope everything goes well,” the founder said as 20 cars drove around the event at Warner Brother Studios with no people inside. “We'll find out!

He added that he expects the robotaxis to be implemented “before 2027” at a cost of less than $30,000.

Convincing regulators and passengers of the vehicle's safety could prove much more difficult and take much longer – while the company's main competitors, such as Alphabet-owned Waymo, are expanding the fleet of robots already operating in select cities.

Tesla has so far had to follow a different technology path than all of its major autonomous vehicle rivals, according to Reuters interviews with more than a dozen executives, consultants and academics – a path that offers potentially higher benefits but also higher risks for both its business and its passengers. specializing in autonomous vehicle technology and three former Tesla autonomous vehicle engineers.

Tesla's strategy relies solely on combining “computer vision,” which aims to use cameras in the same way humans use their eyes, with artificial intelligence technology called end-to-end machine learning, which instantly translates images into management decisions.

The technology already underpins the driver-assistance feature of “fully autonomous driving,” which, despite its name, cannot be safely operated without a driver. Musk said Tesla is taking the same approach to developing fully autonomous robotics.

Tesla's competitors – including Waymo, Amazon's Zoox, General Motors Cruise and a slew of Chinese companies – use the same technology, but typically layer on redundant systems and sensors such as radar, lidar and advanced mapping to ensure safety and obtain regulatory approval. regulatory requirements for their autonomous vehicles.

Tesla's strategy is simpler and much cheaper, but it has two critical weaknesses, industry executives, autonomous vehicle experts and one Tesla engineer told Reuters. Without the layered technologies used by competitors, Tesla's system struggles more with so-called “edge cases” – rare driving scenarios that autonomous systems and their engineers have difficulty predicting.

The second major challenge: The end-to-end AI technology is a “black box,” a Tesla engineer said, making it “almost impossible” “to see what went wrong when it misbehaves and causes an accident.” The inability to precisely identify such failures, he said, makes it difficult to protect against them.

Tesla did not respond to a request for comment on its technology.

Nvidia founder and CEO Jensen Huang used the same “black box” description in an interview to describe the weaknesses of the end-to-end technology, without specifically referring to Tesla's system. End-to-end AI involves training a computer to make decisions directly from raw data, with no intermediate steps that require additional engineering or programming.

Nvidia, the world's leading maker of AI computing chips, also uses end-to-end technology in the autonomous driving systems it is developing, which it plans to sell to car manufacturers. However, Huang told Reuters that Nvidia is combining this approach with more conventional computing systems and additional sensors such as radar and lidar.

Comprehensive technology usually – but not always – makes the best decisions, Huang said, which is why Nvidia is taking a more conservative approach. We have to build the future step by step,” he said. “We can't go directly into the future. It's too dangerous.

Robotaxi axis

Tesla's ability to deliver roboboxes has taken on greater importance this year as sales and profits have fallen amid declining demand for electric vehicles around the world and fierce competition from China's growing electric vehicle makers.

If Tesla can overcome the technical challenges of its autonomous strategy, the benefits could be enormous. While competitors like Waymo already have robots on the road, they operate much more expensive vehicles in relatively small, comprehensively mapped zones.

Tesla's goal is to sell an inexpensive robotic taxi that can drive itself anywhere.

Musk has a long history of making bold promises about autonomous cars. In 2016, he predicted that within two years drivers would be able to hail their vehicles from all over the country. In 2019, Musk predicted that Tesla would produce working robotics by 2020.

This week's robotaxi reveal announcement was made on April 5. Reuters exclusively reported that Tesla had abandoned plans to build a $25,000 electric vehicle for the masses, known informally as the Model 2, which initially sent Tesla's shares tumbling. Musk responded to the appointment later that day on his X social media platform: “Robotaxi unveiling on August 8,” sparking intense speculation among investors. Tesla later delayed the event until this week.

That April day marked a fundamental shift in Musk's stated priorities. He previously pledged to make Tesla an electric vehicle giant the size of Toyota, which underpinned Tesla's soaring stock price, making it the world's most valuable automaker. Now it has vowed to dominate autonomous vehicle technology.

There have been sudden cost-cutting measures, including mass layoffs, as Musk has diverted investment away from electric vehicle production priorities such as battery development, gigacasting and the development of a car factory supercharging network.

The retreat from mass electric vehicles has only increased investor pressure on the development of Tesla's autonomous vehicles. Musk agreed with this analysis and in April said that anyone who doubts that Tesla will “solve the autonomy issue” should not invest in the company.

Nicholas Mersch, portfolio manager at Purpose Investments, a Tesla investor, said Musk “has a lot of convincing to do.”

Still, Mersch called Musk's autonomy strategy “a really bold bet” with a potentially huge payoff, even if it takes Tesla much longer to crack the code. “You have to keep the bigger picture in mind about how much iterative innovation is happening,” he said at Tesla. “I wouldn't underestimate them.”

Data-driven

For now, unlike its competition in the form of Robotxi, Tesla only offers semi-autonomous solutions under the “Autopilot” and “Full Autonomous Driving” features. The naming and marketing of these systems has sparked investigations and lawsuits over whether Tesla put drivers at risk by exaggerating its vehicles' autonomous driving capabilities.

A National Highway Traffic Safety Administration (NHTSA) investigation published in April found that between January 2018 and August 2023, there were 542 crashes, including 14 fatalities, in Tesla vehicles with Autopilot or FSD engaged.

But introducing Autopilot and FSD to mass-produced models gives Tesla a clear competitive advantage: a huge trove of data collected by cameras mounted in millions of vehicles, which it can analyze and use to develop self-driving technology.

Two former Tesla engineers said the technology's relatively low cost enables data collection on a massive scale compared to the relatively small fleets of competitors such as Waymo. One engineer said Tesla's high-resolution cameras cost much less than lidar and could eventually enable the automaker to produce fully autonomous vehicles that customers can afford.

Lidar uses lasers to create three-dimensional images of the vehicle's surroundings while avoiding obstacles.

In a call with analysts and investors this summer, Musk boasted of “exponential” improvement and predicted that Tesla could achieve unattended driving capability “by the end of this year,” adding that he would be “shocked if we can't do it next year.” year”.

Sasha Ostojic — a former self-driving engineer and director of software development at Nvidia, Cruise and Zoox — said he thinks it will take Tesla at least “more than three years” to match the level of autonomous driving currently achieved by Waymo. Ostojic currently advises Palo Alto venture capital firm Playground Global on technology investments.

“I don't see Tesla moving toward true, eye-and-brain autonomous driving,” he said, “on the timeline that Elon Musk promised.”

With postal wireS