Nobody is paying much attention, and that’s exactly the point.
While Silicon Valley startups and a handful of crosstown rivals have spent years chasing the splashy promise of robotaxis — fleets of driverless cars ferrying strangers across San Francisco at 2 a.m. — GM has been quietly playing a longer, stranger game. The company wants to put autonomous driving technology not in a commercial fleet, but in the vehicle sitting in your garage. All of them. Across every brand it sells, on every kind of road, in every kind of weather.
This month, GM entered what it calls the next phase of that effort, beginning supervised on-road testing of its latest automated driving systems across select states. The milestone is unglamorous by tech-world standards. There are safety drivers. There are engineers with laptops. But inside the company, it represents a turning point years in the making.
“Real-world supervised testing helps close the gap between simulation and reality,” the company said in materials outlining its approach — a phrase that sounds modest until you consider the gap it’s trying to close.
The cars are encountering things no simulation can perfectly recreate: an unmarked construction detour at dusk, a lane suddenly washed out by rain, a cyclist who doesn’t signal. Every edge case, every near-miss, every moment of ambiguity gets logged, analyzed, and fed back into an AI system that is, in theory, getting smarter with each mile.
Nearly 700,000 GM vehicles equipped with Super Cruise — the company’s hands-free highway driving technology — have collectively logged more than 800 million miles across 23 models in North America. That is not a prototype. That is a product. And every mile generates data that GM says is accelerating the intelligence of its autonomy stack faster than a smaller company with a smaller fleet ever could.
“A small autonomous fleet might accumulate a few million miles annually,” the company notes. A global automaker deploying driver-assistance systems across its lineup accumulates those miles at a fundamentally different scale.
The practical target of all this is the 2028 Cadillac Escalade IQ, which GM has announced will be the first vehicle to offer what it calls “eyes-off” driving — meaning the driver can fully disengage and let the car handle itself in certain conditions. That is a significant jump from the current state, which requires a driver to keep their eyes on the road even when their hands aren’t on the wheel.
Getting there, GM argues, is as much an engineering problem as a data problem. The company has invested heavily in simulation infrastructure that it claims can model roughly 100 years of human driving in a single day — replaying real events, generating synthetic scenarios, and training AI systems on hazards that might appear once in a lifetime on actual roads but need to be handled reliably every time.
None of this would be possible, the company is quick to point out, without the 2016 acquisition of Cruise, the San Francisco-based autonomous vehicle startup that GM purchased and eventually folded into its own engineering operations. That integration gave the automaker access to sophisticated autonomy expertise that it has spent the years since weaving together with its own driver-assistance technology, hardware development, and manufacturing muscle.
It is a story GM has been trying to tell for a while — though not always to its benefit. Cruise suffered a damaging setback in late 2023 when one of its robotaxis struck a pedestrian in San Francisco, triggering a regulatory suspension of its California driverless permit and a cascading crisis that effectively ended its commercial ride-hailing ambitions. The company laid off hundreds of employees and pulled back sharply on its autonomous fleet.
What emerged from that reckoning, at least in GM’s telling, is a refocused strategy. Less robotaxi. More personal vehicle. The goal is no longer to compete with Waymo for the right to drive strangers around; it’s to give owners of a Cadillac Escalade or a Chevrolet pickup the option to let the car take over.
The economics are different too. GM argues that building autonomy for personal vehicles, rather than specialized commercial platforms, allows it to spread the cost of sensors, compute hardware, and software development across millions of vehicles — driving down the price per unit in a way that companies operating smaller fleets simply cannot replicate. The company builds a vehicle every 60 seconds. That kind of scale changes what’s financially possible.
GM’s history with autonomy stretches further back than most people realize. In the 1950s, its Firebird concept cars imagined vehicles capable of sensing their environment. By the 1990s, GM research vehicles were navigating Southern California freeways autonomously using magnetic road markers — technology that, for its time, was genuinely radical. In 2007, a modified Chevrolet Tahoe named Boss, developed in partnership with Carnegie Mellon University, completed the DARPA Urban Challenge without human intervention, finishing first in a field of sophisticated competitors.