Forget fully autonomous robotaxis: the route to market for driverless technologies now looks like it could be incremental safety autonomy in private cars, a road carmakers have already been travelling for decades.
Yesterday, GM said it would no longer fund the Cruise robotaxi experiment, after buying the driverless company in 2016. Cruise’s robotaxis had been on San Franciscan roads up until recently, following an incident in which a pedestrian was knocked under a Cruise car and dragged to the side of the street.
Instead, the carmaker said it would bring the technology in-house in order to prioritize the development of advanced driver assistance systems, in the hopes that’s the path to fully autonomous cars.
Of course, that still leaves Waymo (owned by Google/Alphabet), smaller startups like May Mobility, and Chinese rivals in the ring with robotaxis. And Tesla last month unveiled the Cybercab, its take on the self-driving robotaxi, expanding its driverless efforts beyond private cars. The vehicle, according to Elon Musk at the time, would go into production in 2026 and cost just $30,000, but investors weren’t convinced and share prices slipped.
But it’s increasingly clear that carmarkers like GM and Ford appear to think robotaxis will take too long to arrive, so they may as well start making money on advanced autonomy features now instead.
Missing business model for robotaxis
Jack Stilgoe, a professor at University College London, says the move was predictable because Ford did the same thing, acquiring Argo AI in 2017 and then closing it down in 2022.
At the time, Ford’s CFO John Lawler said it was “taking that investment and putting it towards a business where we think we will have a sizable return in the near term relative to one that’s going to have a long arc.” In other words, robotaxis aren’t going to pay off anytime soon — perhaps no surprise given the costs of development and kitting out a car versus the much cheaper business model of Uber or Lyft.
“Robotaxis get a lot of attention, but nobody has worked out how they work as a business model,” Stilgoe tells me over email. “On the other hand, car companies know how to sell upgrades and widgets to their richest customers.”
Saber Fallah, professor of Safe AI and Autonomy at the University of Surrey, also wasn’t surprised by GM’s move. “The decision to drop funding for Cruise may also signal a strategic shift, recognising that the technology’s readiness level and return on investment timelines do not align with current business priorities,” Fallah told me in an email. “This underscores the need for a more measured approach, focusing on solving fundamental technical and ethical challenges before pursuing large-scale deployment.”
Driverless challenges may scupper robotaxis
And there are plenty of challenges to overcome — not just for GM and Cruise, but for the wider driverless development industry.
“It reflects the broader reality that the current state of robotaxi technology is not yet mature enough for widespread, safe deployment,” Fallah says. “Despite significant advances in autonomous vehicle development, challenges such as ensuring robust performance in complex urban environments, managing edge cases, and meeting rigorous safety standards remain unresolved.”
As advanced as current AI systems feel, they remain limited in their capability when compared to us humans, Fallah argues.
“While AI systems have made remarkable strides, they still lack critical human cognitive capabilities essential for navigating complex and dynamic environments alongside human road users,” Fallah says.
“Humans excel in reasoning through difficult or ambiguous driving situations, adapting based on experience, and exercising nuanced judgment,” he adds. “Tasks that are trivial for humans — such as interpreting subtle gestures, responding to unpredictable behaviour, or adapting to rapidly changing conditions — remain profoundly challenging for AI.”
Small steps to autonomy beyond robotaxis
When or if such challenges can be overcome — and clearly Waymo et al still believe it’s possible — there are different routes to take.
Those with plenty of funds to burn through can keep at driverless development until full driverless is reached, the as yet to be achieved “level 5” system — or something close enough to it with the added assistance of remote operators stepping in to handle extreme edge cases; that in theory could allow robotaxis. But Fallah and Stilgoe agree that full level 5 autonomy remains an aspiration rather than a reality.
Setting aside robotaxis, carmakers can bring driverless technologies to cars without having the full technology in place by keeping humans in the loop and/or constraining the environment.
The former will see automation methodically added into private vehicles as safety features with humans still at the wheel — as has been happening with existing safety systems for decades. Indeed, driverless research by Ernst Dickmanns in Germany in the 1980s and 1990s didn’t lead to level 5 autonomy, but did contribute to the advanced driver assistance systems safety tools used in cars today.
“Advanced driver assistance systems (ADAS), such as adaptive cruise control, automated parking, and highway assist, represent incremental progress toward autonomy while enhancing safety and convenience for human drivers,” Fallah adds.
The latter would limit autonomous vehicles to controlled environments, known as the operational design domain (ODD) — specific conditions where such a vehicle could be allowed to operate.
“For example, self-driving shuttles and buses operating within dedicated lanes or predefined urban areas are already being tested and rolled out in pilot programmes,” he says. “These systems can function effectively within a limited ODD where the complexity of the environment and unpredictability are reduced.”
That’s already happening: smart safety systems are already in use in newer cars, allowing drivers to hand lane keeping and speed management over to the car. But drivers must remain aware — that’s the “human in the loop” — and such systems only work on motorways, a constrained environment.
The only real difference between ADAS and level-2 driverless in the UK for drivers is that in the former they must keep their hands on the wheel, and in the latter they can rest hands away from the wheel — for both, the driver must remain attentive.
Try everything, see what sticks
Fallah says a mix of all three approaches still seems likely. “Robotaxis, operating within specific geofenced areas, may also play a role, but their success will depend heavily on overcoming technical, regulatory, and trust-related barriers,” he tells me. “For now, the path forward will likely be a combination of these approaches, with autonomy arriving in bits and pieces rather than a single leap to fully self-driving vehicles.”
That suggests GM has made the right move, limiting its spend on robotaxis when that bet isn’t looking likely to pay off anytime soon, while continuing to work on the technology for private cars.
But Stilgoe notes that more questions remain than answers when it comes to driverless, regardless of the route to market — after all, what’s the point of driverless cars anyway?
“Getting a workable self-driving vehicle was always going to be a longer and bumpier journey than the tech bros predicted,” adds Stilgoe. “And we still don’t know how these companies are going to make money from the technology or how they’re going to scale it up to new places and new countries. Nor do we know what transport problems it will solve. We still don’t have anything approaching a trustworthy driving test for self-driving vehicles. At the moment, governments are letting the companies mark their own homework and hoping there aren’t any big issues.”
He adds: “There’s a lot of work still to do before the technology can demonstrate its benefits.”
