Illustration by Fernando Capeto for Forbes; Photos by Apu Gomes/Getty Images; JannHuizenga/Getty Images
Elon Musk seeks to grab eyeballs and headlines with a splashy event Thursday that’s designed to impress investors but is light on safety and testing details. Experts say Tesla is far behind Alphabet’s Waymo, which is quickly growing its commercial robotaxi service.
By Alan Ohnsman, Forbes Staff
E lon Musk’s much-hyped Tesla “CyberCab” show tonight at Warner Brothers Studios in Burbank, California, the billionaire CEO is expected to announce a new electric model that will be the basis for Tesla’s futuristic taxi fleet. Musk also will likely use the invitation-only event – dubbed “We, Robot” – to tout the revenue-generating potential of such a business for Tesla and share a launch date.
His desire to excite investors and Tesla fans may lift its stock price, which is flat this year on lower EV sales. But there’s not much public information about Tesla’s autonomous driving progress. It stopped filing reports with California’s DMV, one of the main sources, in 2019 and it doesn’t brief outside experts or the media on its progress. There’s also ongoing scrutiny of its Autopilot and Full Self-Driving software by regulators, and known, ongoing technical problems with these suggest attaining Musk’s vision anytime soon would be a stretch.
“Tesla is not a serious contender for real self-driving applications,” said George Mason University professor Missy Cummings, an artificial intelligence expert who served as an advisor to NHTSA on autonomous vehicles and consults California’s Department of Motor Vehicles on the topic. “I think they are, at a minimum, 10 years away from such operations.”
Hollywood stunts aside, the most promising robotaxi program right now isn’t Tesla’s CyberCab but Waymo’s, which has long been the self-driving industry benchmark. Fifteen years after it began as the Google Self-Driving Car project, nine years since its first fully autonomous test drive ferrying a vision-impaired passenger through Austin in 2015, and aided by at least $10 billion of investment, Waymo has built an actual business, though far from a profitable one.
“At the end of the day, this is a huge opportunity”
Its robotaxis, with no humans at the wheel, currently handle more than 100,000 paid rides a week in Phoenix, San Francisco and Los Angeles. That number is likely to swell soon as the service expands to Austin and Atlanta. The vehicles aren’t flawless, exhibiting odd behavior like inexplicably honking in a Waymo parking lot, freaking out in front of Kamala Harris’s motorcade or occasionally causing minor traffic jams when blocked by double-parked cars and trucks. Waymo has avoided the major accidents that stymied rival programs at Uber and Cruise, as well as the crashes and fatal accidents linked to Tesla’s partially automated software, though NHTSA is investigating several collisions and traffic violations involving its robotaxis (these instances didn’t result in injuries or fatalities).
Dmitri Dolgov, the Alphabet venture’s co-CEO who’s worked to solve autonomy since Google launched the effort in 2009, is trying to keep it that way.
“In everything we do we approach this with safety as the primary principle,” he told Forbes. “That extends to the scaling [of the service] that you see now. It means that we continually raise the bar on the performance of the system, on the rigor of our evaluation and the validation and the readiness framework. All of those things evolve, and that’s how we’ll keep approaching it.”
Bullish estimates for a future robotaxi market are $50 billion a year by 2030, according to a recent Raymond James report. Forbes estimates Waymo’s revenue will top $50 million this year, potentially approaching $100 million as the service expands, but Dolgov won’t confirm this or discuss when the service will be profitable. He won’t discuss Tesla, either.
“At the end of the day, this is a huge opportunity and it’s great that this field attracts some super brilliant minds and resources,” he told Forbes. “There’s room for different approaches, and at the end of the day if one can solve this problem and it accelerates the mission and the vision that’s great.”
Human Monitors, Multiple Sensors
Waymo and Tesla have followed vastly different paths to solve autonomous driving. Waymo has conducted years of structured testing, at its own track in California’s Central Valley, and monitored testing in cities, in both metro Phoenix and Silicon Valley. It’s worked to master specific driving maneuvers that humans take for granted, like making unprotected left turns and navigating around obstacles in the road, by doing them countless times, as well as teaching its AI to recognize street signs, traffic cones, emergency vehicles and tell the difference between plastic bags in the street and hard objects. It started with slower, easier-to-maneuver suburban streets, with fewer pedestrians and lower speeds, before turning its attention to dense urban environments, like San Francisco.
Waymo autonomous vehicles parked in a staging area in San Francisco
Getty Images
Waymo’s system also relies on remote human monitors who keep tabs on its robotaxis, providing guidance to vehicles if they’re stumped by road conditions, interacting with police in the event of a mishap or just answering questions for riders. That adds to the cost of its operations but is key to Dolgov’s goal of ensuring safety, he said.
Waymo also uses far more robust sensors than Tesla to ensure its vehicles can see and understand as much of the world as possible, including laser lidar for 3D images in daylight or darkness, radar, cameras and audio sensors (important for hearing emergency vehicles).
Tesla, by contrast, is collecting real-world driving data from its customers via a neural network with the goal of training its software to handle any of the road and weather conditions drivers experience. And for Tesla passenger vehicles, it uses only cameras as a sensor. It’s a much lower-cost approach, but one that’s a concern for AV researchers.
“Any form of autonomous control is only as good as its sensory input,” said Bart Selman, a professor of computer science and engineering at Cornell University, noting that the Boeing 737 MAX 800 accidents are an example of what can happen if you have a malfunctioning sensor and don’t build in any redundancy. “I would be surprised if Tesla could provide reliable, fully autonomous driving with just camera input. It may appear to work well most of the time, but it does not seem sufficiently robust and safe for broad deployment.”
“As exciting as the robotaxi concept is, investors expecting it to alter Tesla’s financials in the short term will be disappointed”
Tesla’s current automated software, a $15,000 option sold to customers as Autopilot and FSD, or Full Self-Driving, has been highly controversial, never living up to Musk’s hype. A detailed review of crashes by the Wall Street Journal, including several that resulted in fatalities, found that Tesla’s system struggles to recognize objects and obstacles on the road. The lack of additional sensors beyond cameras appears key to that problem. While Tesla’s system warns drivers when they need to take control of the vehicle, it’s possible that in many cases this doesn’t happen in time to avoid a collision. California, Tesla’s top market, is suing the company for “misleading and false advertising” over the use of the names Autopilot and Full Self-Driving as they’re likely to mislead consumers about their capabilities.
Unmet Promises
Musk has a habit of making bold promises and not following through on them, from plans for hyperloops and solar roofs (unveiled at Universal Studios, next door to Warner Brothers in 2016) to selling 20 million EVs a year by the end of the decade. That extends to robotaxis: In 2019, he pledged that Tesla could have a million robotaxis operating by 2020. Since then, he’s clearly fixated on pivoting Tesla into a robotaxi powerhouse, despite the fact that 90% of its revenue comes from selling electric cars, the pollution credits they generate and batteries. That’s because it looks like a high-margin business – if only the AI-enabled driving system can be perfected.
Teslare recalled more than 2 million vehicles to fix a defective system that’s supposed to ensure drivers pay attention when using Autopilot.
AP Photo/Noah Berger, File
“The value of Tesla overwhelmingly is autonomy,” Musk told shareholders in Tesla’s July earnings call. “So I recommend anyone who doesn’t believe that Tesla would solve vehicle autonomy should not hold Tesla stock. They should sell their Tesla stock.”
How long it will take Tesla to do what Musk promises isn’t clear.
“As exciting as the robotaxi concept is, investors expecting it to alter Tesla’s financials in the short term will be disappointed,” said Karl Brauer, a long-time industry analyst for the consultancy iSeeCars. “Getting self-driving cars through the regulatory hurdles and onto the street in meaningful numbers will take time, even if Elon were to deem the technology ready for public consumption this week.”
Alphabet also doesn’t break out Waymo’s revenue from its Other Bets business, which totaled $365 million in the year’s second quarter, up 28%, or discuss its expectations for the venture, though CEO Sundar Pichai is upbeat.
“I’m really pleased with the progress Waymo is making, a real leader in the space and getting rave reviews from users,” he said on Alphabet’s earnings call in July.
George Mason’s Cummings, dismissive of Tesla’s program, sees Waymo’s efforts as more thorough but remains skeptical that any company can fully master autonomous driving in the near term.
A review of test data submitted to regulators by Waymo, Cruise and Zoox shows autonomous vehicles “can drive in a way that is generally technically legal, but occasionally severely violates ‘norms of the road’ that are expected by human drivers,” she wrote in a draft of a research paper going through peer review shared with Forbes. (Tesla is not included in the analysis as it doesn’t share data regarding “disengagements” – situations when the autonomous system is overridden by a test driver.)
Some accidents reported, “could be the fault of other drivers (like a truck backing into the AV at a turn), however, humans expect a certain type of defensiveness and cooperation in the settings where these events are occurring and the AVs do not consistently exhibit these abilities,” she wrote.
Whether it’s Tesla or Waymo, researchers such as the University of Michigan’s Henry Liu, who runs a test facility for autonomous vehicles in Ann Arbor, think the federal government should create a driving test specifically for them before they can use public roads.
Ultimately, Dolgov has his priorities: reduce the costs of its system and don’t blow the lead Waymo has built by expanding too fast. The first point is being addressed with a so-called sixth generation of sensors, computing system and software that will be used on lower-priced electric vehicles in Waymo’s fleet next year, including Hyundai’s Ioniq 5 hatchback.
“We’ve been very intentional about our scaling and deployment,” he said. “We’re intentionally scaling to allow the Waymo driver to learn as quickly as possible and to learn from our customers, learn from the experiences and put that forward.”
Still, growth is coming. “Given where we are today, the scale that we’re operating at today and the diversity of the operating environments, you can find a lot of zip codes where the environment is completely within the operating domain of what we have today,” he said.