General Motors completely redesigned the compute architecture that powers its next-generation “hands-free” driving system thanks to US semiconductor company Qualcomm.
The automaker’s Ultra Cruise system, which will make its debut in the 2024 Cadillac Celestiq electric sedan, will be the first advanced driver assist system (ADAS) to use Qualcomm’s new Snapdragon Ride Platform. The size of “two laptops stacked together,” GM claims the new architecture will have the processing power equivalent to “several hundred” personal computers.
GM first announced Ultra Cruise during an investor event last year, describing it as a massive leap over the company’s Super Cruise system, which allows for hands-free driving on mapped, divided highways. In contrast, Ultra Cruise will cover “95 percent” of driving scenarios on 2 million miles of roads in the US, the company claimed.
“We’re attempting to have this feature be sort of a door-to-door driverless operation,” said Jason Ditman, chief engineer at GM, in an interview with The Verge. “When the vehicle gets onto a capable road, Ultra Cruise will automatically engage and handle the majority of the work, hands-free. Stop signs, stoplights, turns, splits, merges, freeways, subdivision… all of those domains.”
That’s thanks to Qualcomm’s new high-powered processors, Ditman said. Last year, Qualcomm entered into an agreement with GM to provide computer chips for the automaker’s next generation of electric vehicles. When it comes out in 2023, the Cadillac Celestiq will be one of the first vehicles to feature the chipmaker’s new ADAS platform, which includes Qualcomm’s Snapdragon SA8540P system-on-a-chip and SA9000P artificial intelligence accelerator.
Those two SoCs will deliver “low-latency control functions on 16-core CPUs and high-performance AI compute of more than 300 Tera Operations Per Second for camera, radar, and lidar processing,” GM says. But more importantly, it saves the automaker a lot of room.
“It lets us be real efficient with space,” Ditman said. “It gives us the throughput that we need. And it let us take what was originally architected to be distributed modules throughout the car and move all that functionality into the one box.”
Consolidating modules for driver monitoring, mapping, and video processing into one box has also improved the system’s latency, Ditman said, because it allows GM to bypass the previous system routed through the vehicle’s CAN network. Ultra Cruise will use sensors like cameras, radars, and lidar to control the vehicle’s steering, acceleration, and braking. GM hasn’t announced all the details around the sensor suite, but Ditman noted that the system relies on a total of 20 different devices.
Despite its enhanced capabilities, GM says it still considers Ultra Cruise a Level 2 system, as defined by the Society of Automotive Engineers. At Level 2, the vehicle can control both steering and acceleration and deceleration, as well as monitor blind spots and even change lanes automatically. But the driver needs to stay vigilant and keep their eyes on the road; if they don’t, the vehicle’s infrared sensors will detect it, and the system will send several warnings to the driver before disengaging.
Ditman said GM learned a lot about driver behavior from monitoring their behavior and made appropriate modifications for Ultra Cruise based on those findings. Some studies have shown that drivers are slow to take back control from a Level 2 system, a troubling sign that could undermine the safety case for more advanced driver assistance. But Ditman said GM’s robust driver monitoring system serves as a “good training tool to get people to pay attention.”