GM’s ‘Ultra Cruise’ Tries To Be Tesla FSD ‘Beta’ Done Right, But Should It?

General Motors has announced that their new “Ultra Cruise” product will be available (eventually) on the Cadillac Cellistiq. The Cellistiq will go on sale in early 2024 with the necessary hardware, but the Ultra Cruise will be operational only when it’s ready and safe, says Jason Ditman, chief engineer on the project.

Unlike GM’s “Super Cruise” which controls the car while you watch on freeways, the Ultra product will do this in 95% of all roads and driving situations, says GM. In other words this has many similarities to the so-called “beta” version of Tesla’s badly named Full Self-Driving system, which today is operating only as a driver-assist system which requires constant driver attention, though it promises (and keeps promising every year) to someday drive on its own.

GM released the details on their sensor suite, which is more full than Tesla’s, with 20 sensors, a 360 degree wrap-around camera view from 7 cameras, 4 short-range radars, long range radars and a LIDAR near the rear-view mirror. Tesla has 8 cameras and nothing else, though the next generation Teslas will feature at least 11 cameras and possibly a new radar (even though Tesla declared radar to have negative value last year.) GM also strongly promotes their driver monitoring system, which has a camera to watch the driver’s eyes and make sure they stay on the road. Tesla monitors torque on the wheel, and more recently also started using a small camera in the rear-view mirror, though its resolution and angle are limited.

Organizations like Consumer Reports have been very keen on good driver monitoring; they have gone so far as to downrate what I call “Driver Assist Pilot” (DAP) systems if they don’t nag the driver enough to pay attention, or allow drivers to activate the systems in anything but the safest areas to do so. While it’s doubtful that consumers are actually saying, “I don’t want to buy that one, it doesn’t force me enough to stay safe,” CR is quite firm on this point. (It’s not that the feature is not valued, but rather that the actual quality of the driving carries almost all the weight in a consumer’s valuation of these systems, and the amount of nagging is relatively low.)

GM’s improved sensor suite hopes to deliver superior driving performance, though. However, surprisingly for a system they hope to ship in a year, they have yet to test it on public roads. This means that they don’t have any real world user experience, other than what has been learned from Super Cruise (which Ditman also managed.)

Indeed, Ditman was unable to answer the key question of “Will customers actually like it?” While many customers like both Tesla Autopilot and GM Super Cruise (the latter even more as it does not demand hands stay on the wheel) many Tesla drivers who try out the FSD “beta” do not find it relaxing or pleasant. Highway driving is very simple, and easy to supervise City driving involves sharp turns and close encounters with other vehicles, many moments where the difference between a safe action by the car and a disaster are not easy to tell apart. You can’t keep your hands easily on the rapidly turning wheel — again, something the GM system will do better — but you also can’t relax. Most FSD owners like experimenting with it but only a fraction say it is a pleasant and relaxing drive.

Ditman’s goal, he says, is to make sure drivers are “confident” that the system is good, so that they can relax. It is an open question of how confident we want drivers of such systems to be. Confidence triggers “automation complacency” which is why driver monitoring is a big deal, but it’s possible to zone out or become less diligent even with eyes on the road and this may be exactly where you don’t want much confidence. It is unclear whether hands-free systems are better or worse than “hands on the wheel with low torque” in this regard.

Tesla recently published that FSD is getting one airbag-deploying crash every 3.2 million miles. That’s twice as many as Autopilot, but Autopilot goes on highways which have a much lower crash rate than city streets, and FSD only operates on city streets. It is reasonable to posit that Tesla’s number is exactly because FSD is a poor system that needs regular intervention. Anybody who gets complacent and tries to let FSD do the job while they ignore the road will crash within a short time, and as such, people don’t ignore the road.

Ditman stated he didn’t know how many crashes Super Cruise users have had, because he no longer heads that project. In early operation, GM reported there were no crashes. There have been many crashes with Tesla Autopilot, including fatalities, largely with drivers who ignored the rules and ignored the road.

To be uncharitable, it seems very strange that the former head of Super Cruise, who has expressed that safety is his top priority would be unaware of whether there were crashes with his former baby. As he develops Ultra Cruise, I would expect him to study any crashes in that system in great detail to assist with the design of the new system. GM stated they would provide the statistics, but did not by press time. (Check for update later in the day.)

GM is including a “heads up display” with Ultra Cruise in the Cellistiq, but declined to indicate what it would display. It is not appropriate for drivers to spend much time looking at the dashboard display, so the HUD will be crucial. A HUD which shows the planned path of the system projected onto the real world, so the driver is never surprised, and can see if that path ever will take the car somewhere it should not go could be an excellent tool in making the system safer, and the task more relaxing. With such a HUD, you would mainly watch the road, and watch your future path through it, and intervene if you don’t like what you see.

Ultra Cruise will avoid any tricky sections of road, including all roundabouts and very complex intersections. As you approach these, you will get a signal to take control. They are putting lights in the wheel and dash to make sure the driver is aware of the state of the system and what they must do. This won’t work very well in Europe, where there’s a roundabout every 30 seconds in some areas. The system will also need you to take over before entering driveways or parking lots. On the other hand, it can be argued that regular brief sections where the driver must take over will keep drivers on their toes. I and others have even suggested that such systems deliberately get close to a lane departure without warning when the road is clear and it is safe to change lanes, just to make sure the driver is watching. You don’t want them to get too confident.

GM had no extra comments on their LIDAR, or on the role of mapping. They said that this system was developed independently of their Cruise robotaxi subsidiary, in spite of the name (which predates Cruise the company) and in spite of the public fight where Mary Barra fired the CEO of Cruise over a desire that the subsidiary help the parent more with products like this.