The race towards a driverless future continues with news over the past few days that GM, America’s biggest carmaker by sales, is investing $35 billion in developing electric cars and driverless technology in the run-up to 2025. Waymo, the Google spin-out that is considered to be furthest ahead in its development of road-tested driverless technology, has also announced the raise of an additional $2.5 billion in cash from external investors, adding to the $3.2billion round closed in late 2020.
Billions of investor cash continues to be poured into driverless technology, and R&D and road testing carries on in the background (Waymo is even running a pilot commercial scheme for driverless taxis in Phoenix’s East Valley district). But the past year or so has been quiet when it comes to solid news on progress towards driverless vehicles becoming a regular sight on our streets.
That relies on final tweaks to the driverless technology stacks being developed by companies such as GM, Waymo and Tesla, among others. Enough test miles being driven on roads in real life conditions to both demonstrate the technology’s safety and further feed the algorithms that will make split-second decisions on how cars react in countless scenarios. And finally, arguably most importantly, convincing regulators to sign off on the new legislation that will be needed to allow for machine-operated vehicles on the roads.
The latter was always likely to be a bottleneck due to numerous complications, not least around liability in the case of accidents. Legal responsibility and implications for the insurance industry are big questions to solve, and will involve a lot of debate and, inevitably, controversy.
Waymo’s new investment round has also raised new questions around the company’s rate of cash-burn, coming just half a year after its previous multi-billion cash call. The fact that a company owned by Alphabet, which is sitting on a $135 billion cash pile, is bringing outside investors in to help fund Waymo, which is unlikely to generate meaningful revenue for at least another few years, could be telling. Alphabet did, however, lead the funding round.
Waymo’s co-chief executives Dmitri Dolgov and Tekedra Mawakana yesterday commented after the investment round was announced:
“There’s no greater challenge in artificial intelligence than building and deploying fully autonomous technology at scale.”
The huge expense involved in the pursuit of driverless vehicles has seen a wave of consolidations across the industry. Last year saw a merger between Uber’s driverless business and Aurora, while Lyft sold its “Level 5” business to Toyota and GM’s Cruise snapped up driverless start-up Voyage in March of this year.
Getting driverless vehicles on the road in a commercial capacity will take longer than many in the industry had hoped a few years ago. And cash burn rates higher than anticipated in the meanwhile. That may be starting to hit valuations. Waymo was valued at around $30 billion when it raised investment in late 2020. The company declining to comment on the valuation the recent $2.5 billion raise was made it indicates it may have been a down round.
Experts estimate Waymo is burning between $1 billion and $2.5 billion a year paying for the estimated 2400 highly paid staff employed. The fact eight high-level executives have left the company over the past year, including those in charge of finance, manufacturing, investor relations, and safety, may also indicate internal frustrations at the pace of progress.
Perhaps most tellingly, John Krafcik stepped down as Waymo chief executive in April after 5 years at the helm. Krafcik’s professional history is as a production specialist, he was previously in charge of Hyundai’s North American operations, and was brought in to get tens of thousands of robotaxis into operation across multiple cities.
His departure from the role could be a major clue it is accepted that will now take longer than anticipated when he was hired. The profile of Waymo’s two new co-chief executives also offers some insight.
Tekedra Mawakana has a background in regulated technologies, advising consumer technology companies on how to further their business interests and managing growth and risk-mitigation strategies. Dmitri Dogov is one of the founders of the Google self-driving car project which was initiated in 2009 before becoming Waymo in 2016. He previously worked on autonomous driving technology for Toyota and on Stanford’s DARPA Urban Challenge – a 2007 driverless cars competition. Previously Waymo’s CTO, he is a technologist.
The combined skillsets and experience of the new team leading Waymo suggests the company sees navigating and lobbying around regulation and legislation and technology, as its two most pressing immediate challenges. Rather than the operational project of deploying fleets of driverless robotaxis.
Just a year ago driverless vehicles companies were making confident statements that the technology itself was almost ready for commercialisation and regulatory approval and legislative frameworks were now the biggest bottleneck.
But Waymo yesterday said the new funds raised would be spent to “continue advancing” the Waymo Driver operating system it plans to popularise through partnerships with car manufacturers. The intended route to market and presumed business model has many similarities to how Google licenses its Android operating system to smartphone makers.
When will driverless vehicles be on the road?
Five years ago, many experts were predicting we would already have driverless cars and commercial robotaxi services operating on the roads in major cities around the world by 2021. Progress has proved slower than that and other than Waymo’s pilot scheme in the suburbs of Phoenix, Arizona, it looks like it will be another few years before driverless vehicles become a regular fixture on our roads.
A recent Wall Street Journal article quotes Chris Urmson, head of autonomous trucking startup Aurora, which recently acquired Uber’s self-driving division. (Uber also invested $400 million in Aurora), as commenting:
“We’re going to see self-driving vehicles on the road doing useful things in the next couple of years, but for it to become ubiquitous will take time.”
There is now an acceptance that driverless technology will be introduced slowly and in stages. There will be no “green light moment” of lawmakers around the world signing off on driverless technology.
True driverless technology, not an advanced form of cruise control that still requires a human driver ready to take over, will most likely be limited to light vehicles operating in restricted zones. For example, delivery vehicles trundling around suburban neighbourhoods at low speeds, dropping off groceries and other e-commerce orders. Or shuttle services between airport terminals.
Aurora’s driverless trucks, for example, will initially be restricted to operating only along specific motorways for which high resolution, three-dimensional maps have been built. The company says that eventually trucks and cars will be able to use its technology to go driverless away from those artery roads but now refrains from any statements on when that might actually happen. And in the meanwhile, Aurora trucks will still need a driver to take control when they are not driving along designated stretches of motorway.
Some academics are convinced AI still has fundamental shortcomings that mean it’s unclear when, or even if, truly autonomous vehicles will become a reality. Mary Cummings, professor of computer science and director of the Humans and Autonomy Lab at Duke University, has developed a 4-level system to categorise levels of AI used by driverless technologies.
Level 1 – skill-based “bottom-up” reasoning
AIs are already quite good at things like teaching themselves to stay within lines on a road.
Level 2 – rule-based learning and reasoning
This level deals with AI taking decisions like what it should do at a stop sign. Today’s AI can handle that.
Level 3 – knowledge-based reasoning
This level of AI can handle scenarios that involve questions like “is it still a stop sign if half is obscured by a tree branch?”
This is the level at which the AIs driverless vehicles currently run on start to encounter difficulties.
Level 4 – expert reasoning
This involves AI encountering a completely new scenario and having to take a correct decision by applying knowledge, experience ance skills. The kind of split-second decision a driver has to take if a dog, or human, unexpectedly runs across a road and evasive action has to be taken in adverse weather conditions. These scenarios are still a problem for driverless AI.
Dr Cummings says the AI that has been developed for driverless vehicles still can’t properly achieve level 3 but tries to compensate for that by creating ultra-detailed 3D maps, which compensate for gaps in the data the system is fed from sensors on a vehicle.
Machine learning algorithms are fantastic at processing huge volumes of data and finding and matching patterns. But they are weak when it comes to extrapolation – transferring and applying things learned in one area to another. Dr Cummings explains:
“When you’re a toddler, you’re taught the hot stove is hot. But AI isn’t great at transferring the knowledge of one stove to another stove. You have to teach that for every single stove that’s in existence.”
She adds:
“Billions of dollars have been spent in the self-driving industry and they are not going to get what they thought they were going to get.”
She thinks driverless vehicles will be in operation by the end of the decade but in a restricted way that “won’t be what everybody promised.”
However, Waymo’s head of behaviour Nathaniel Fairfield disagrees and sees no fundamental technological barriers to robotaxi services becoming common if regulators and lawmakers allow them to, commenting:
“If you’re overly conservative and you ignore reality, you say it’s going to take 30 years—but it’s just not.”
But if AI just isn’t up to the job, how will that happen? It would involve systems engineering and a huge investment to upgrade roads with transponders and sensors to help guide driverless vehicles and correct them if things go wrong.
That would limit driverless modes to certain roads and probably even weather conditions. Right now, it seems more likely that for the foreseeable future driverless technology will lead to safer human-driven cars and autopilot functionality under certain conditions.
What it probably won’t mean is the impending end of private car-ownership and being picked up and dropped off at the touch of an app by a robotaxi, as the general presumption has been.
The billions still being invested by Waymo, Aurora, Tesla, and others in the pursuit of driverless technology suggests these companies are still confident they are on the verge of something truly groundbreaking. But right now it doesn’t sound like the kind of driverless future we imagined will be with us before the end of the current decade.