By Aamir Ali, associate head of product management, AV and ADAS
By Aamir Ali, associate head of product management, AV, and ADAS
The era of self-driving personal vehicles is getting closer, thanks to recent advancements in sensors, AI software, and hardware. The potential it offers is world changing. Autonomous vehicles can give commuters the gifts of time and peace of mind, while making roads safer.
General Motors has been at the forefront of autonomous driving development for years, and the work is now gaining momentum, as we expand our fleet of vehicles collecting real-world data, adding thousands of miles a day, and millions of miles more in simulations.
Building this technology is a marathon of design, engineering, and validation that requires patience, precision, and a proven path to scale. But we’re getting closer to a future of fully autonomous vehicles, deployed at scale across our portfolio.
Super Cruise shows the way
Building consumer trust in hands-free driving began with GM’s Super Cruise.1 This technology is available on 23 models from all GM nameplates, both EVs and gas-powered vehicles, like the Chevy Equinox, the Cadillac Escalade and the GMC Sierra EV, among many others. Super Cruise operates on over 600,000 miles of mapped roads in North America, with expansion underway for South Korea, the Middle East and Europe.
Super Cruise has driven 545 million hands-free miles with zero reported crashes attributed to the technology. The system enables high-confidence hands-free driving via multi-sensor fusion - cameras, radar and lidar, GPS and maps - all while continuously monitoring driver attentiveness.
Over time GM has added new capabilities to Super Cruise, like hands-free trailering2, integration with Google Maps to show compatible routes, and ensure you’re navigated to the correct lane at exit ramps or interchanges.
Super Cruise is real-world proof that you can deploy complex technology safely and successfully at scale when you take it step-by-step. We're following this same approach as we expand our autonomous driving capabilities and scale it across vehicles and geographies.
Driving GM's journey to self-driving
GM’s acquisition of Cruise earlier this year came with a decade's worth of Level 4 autonomous driving expertise and infrastructure. Cruise has already cracked the hard problems: simulation engines that can run millions of scenarios daily to identify most challenging situations and issues early; multimodal perception systems trained on real as well as synthetic data to get a high confidence 360-degree view of the world; AI platforms that enabled millions of driverless miles; and people who could bring it all together. This powerful combination has provided GM with intellectual property, hardware, software, testing, and AI that's logged millions of road miles.
Testing and simulation
To deliver autonomy you can trust, we are leveraging the expertise from both GM and Cruise -- building our systems using both real-world data and virtual simulations, and rigorously validating our software both in simulations and closed courses.
A GM test fleet equipped with cameras, radars, and lidar is driving across America building the foundation for the next phase of our autonomous journey. Each mile driven generates millions of data points - how shadows fall across lane markers at different times of day, how rain affects sensor readings, how drivers behave in different scenarios like construction zones. This data feeds our AI model, which learns from real-world driving scenarios to prepare for diverse driving conditions, weather patterns, and road types.
Real-world driving data and testing is essential but rarely reveals rare edge cases. Road testing alone, even at millions of miles, cannot surface the diversity or frequency of safety-critical events needed to build and evaluate an AI model with high statistical confidence. Along with being rare, some scenarios, like collisions, are also extremely dangerous and can be nearly impossible to replicate in closed courses.
That's where simulation changes the equation.
GM's simulation framework, developed and refined at Cruise, enables a fully closed-loop evaluation of the autonomy stack, in which we can test in near real-world conditions - interactions between the AI model, the car, actors in the scene and the physical world
Caption: Flow chart showing how data flows through a closed-loop simulation system.
We simulate both sensor physics and human behavior. That means modeling how our sensors see the world and how other drivers and pedestrians react to the autonomous vehicle's decisions. Each day, we run millions of these high-fidelity closed loop simulations, equivalent to more than 10,000 times the daily time spent by an average U.S. driver.
INSERT VIDEO 1 HERE
Caption: This video shows a high-speed highway merge scenario, demonstrating how we simulate sensors and feed it back to the autonomous driving system.
Here's how it works: We stress-test our systems against a diverse set of scenarios drawn from the millions of miles of real-world driving data, national crash databases, and academic research. We can replay actual events, modify real-world data to create new virtual scenarios or design new ones entirely from scratch. This provides us with the ability to have comprehensive coverage of hazardous scenarios and allows us to regularly test the system against rare and hazardous situations that would be nearly impossible to encounter safely in the real world.
INSERT VIDEO 2 here
Caption: This video shows a highway scenario with fog and a late splitting motorcyclist coming up from behind. This example is based on a real-world example, amplified by the addition of fog.
When a potential failure appears in simulation, the system triggers further regression testing, replaying the issue against both its original scenario and every variation we can generate. This process helps us identify issues faster and iterate on our software quickly, and continuously validate its safe performance.
For simulation to be effective and reliable – outcomes and performance in simulation must be predictive of what can happen in the real world. To ensure our simulation systems are valid, we continuously compare the performance in simulation with real-world data to ensure our systems are calibrated correctly and are providing accurate signals for autonomy development and validation.
So, when will GM vehicles drive themselves? Stay tuned. What I can tell you is that our track record suggests we will get there, and our comprehensive development approach ensures we're building that future the right way: step-by-step, mile-by-mile, simulation-by-simulation.
Aamir Ali is associate head of product in GM’s autonomy software product management group. He previously was head of product, simulation, and test at Cruise. Questions or comments: news@gm.com
1Always pay attention while driving and when using Super Cruise. Do not use a hand-held device. Requires active Super Cruise plan or trial. Terms apply.
2Lane Change on Demand and Automatic Lane Change not available while trailering. Always pay attention while driving and when using Super Cruise. Do not use a hand-held device. Requires active Super Cruise plan or trial. Terms apply.
The era of self-driving personal vehicles is getting closer, thanks to recent advancements in sensors, AI software, and hardware. The potential it offers is world changing. Autonomous vehicles can give commuters the gifts of time and peace of mind, while making roads safer.
General Motors has been at the forefront of autonomous driving development for years, and the work is now gaining momentum, as we expand our fleet of vehicles collecting real-world data, adding thousands of miles a day, and millions of miles more in simulations.
Building this technology is a marathon of design, engineering, and validation that requires patience, precision, and a proven path to scale. But we’re getting closer to a future of fully autonomous vehicles, deployed at scale across our portfolio.
Super Cruise shows the way
Building consumer trust in hands-free driving began with GM’s Super Cruise.1 This technology is available on 23 models from all GM nameplates, both EVs and gas-powered vehicles, like the Chevy Equinox, the Cadillac Escalade and the GMC Sierra EV, among many others. Super Cruise operates on over 600,000 miles of mapped roads in North America, with expansion underway for South Korea, the Middle East and Europe.
Super Cruise has driven 545 million hands-free miles with zero reported crashes attributed to the technology. The system enables high-confidence hands-free driving via multi-sensor fusion - cameras, radar and lidar, GPS and maps - all while continuously monitoring driver attentiveness within the vehicle and providing alerts when attention is needed.
Over time GM has added new capabilities to Super Cruise, like hands-free trailering2, integration with Google Maps to show compatible routes, and ensure you’re navigated to the correct lane at exit ramps or interchanges.
Super Cruise is real-world proof that you can deploy complex technology safely and successfully at scale when you take it step-by-step. We're following this same approach as we expand our autonomous driving capabilities and scale it across vehicles and geographies.
Driving GM's journey to self-driving
GM’s acquisition of Cruise earlier this year came with a decade's worth of Level 4 autonomous driving expertise and infrastructure. Cruise has already cracked the hard problems: simulation engines that can run millions of scenarios daily to identify most challenging situations and issues early; multimodal perception systems trained on real as well as synthetic data to get a high confidence 360-degree view of the world; AI platforms that enabled millions of driverless miles; and people who could bring it all together. This powerful combination has provided GM with intellectual property, hardware, software, testing, and AI that's logged millions of road miles.
Testing and simulation
To deliver autonomy you can trust, we are leveraging the expertise from both GM and Cruise -- building our systems using both real-world data and virtual simulations, and rigorously validating our software both in simulations and closed courses.
A GM test fleet equipped with cameras, radars, and lidar is driving across America building the foundation for the next phase of our autonomous journey. Each mile driven generates millions of data points - how shadows fall across lane markers at different times of day, how rain affects sensor readings, how drivers behave in different scenarios like construction zones. This data feeds our AI model, which learns from real-world driving scenarios to prepare for diverse driving conditions, weather patterns, and road types.
Real-world driving data and testing is essential but rarely reveals rare edge cases. Road testing alone, even at millions of miles, cannot surface the diversity or frequency of safety-critical events needed to build and evaluate an AI model with high statistical confidence. Along with being rare, some scenarios, like collisions, are also extremely dangerous and can be nearly impossible to replicate in closed courses.
That's where simulation changes the equation.
GM's simulation framework, developed and refined at Cruise, enables a fully closed-loop evaluation of the autonomy stack, in which we can test in near real-world conditions - interactions between the AI model, the car, actors in the scene and the physical world.
We simulate both sensor physics and human behavior. That means modeling how our sensors see the world and how other drivers and pedestrians react to the autonomous vehicle's decisions. Each day, we run millions of these high-fidelity closed loop simulations, equivalent to more than 10,000 times the daily time spent by an average U.S. driver.
This video shows a high-speed highway merge scenario, demonstrating how we simulate sensors and feed it back to the autonomous driving system.
Here's how it works: We stress-test our systems against a diverse set of scenarios drawn from the millions of miles of real-world driving data, national crash databases, and academic research. We can replay actual events, modify real-world data to create new virtual scenarios or design new ones entirely from scratch. This provides us with the ability to have comprehensive coverage of hazardous scenarios and allows us to regularly test the system against rare and hazardous situations that would be nearly impossible to encounter safely in the real world.
This video shows a highway scenario with fog and a lane splitting motorcyclist coming up from behind. This example is based on a real-world example, amplified by the addition of fog.
When a potential failure appears in simulation, the system triggers further regression testing, replaying the issue against both its original scenario and every variation we can generate. This process helps us identify issues faster and iterate on our software quickly, and continuously validate its safe performance.
For simulation to be effective and reliable – outcomes and performance in simulation must be predictive of what can happen in the real world. To ensure our simulation systems are valid, we continuously compare the performance in simulation with real-world data to ensure our systems are calibrated correctly and are providing accurate signals for autonomy development and validation.
So, when will GM vehicles drive themselves? Stay tuned. What I can tell you is that our track record suggests we will get there, and our comprehensive development approach ensures we're building that future the right way: step-by-step, mile-by-mile, simulation-by-simulation.
Aamir Ali is associate head of product in GM’s autonomy software product management group. He previously was head of product, simulation, and test at Cruise. Questions or comments: news@gm.com
1Always pay attention while driving and when using Super Cruise. Do not use a hand-held device. Requires active Super Cruise plan or trial. Terms apply.
2Lane Change on Demand and Automatic Lane Change not available while trailering. Always pay attention while driving and when using Super Cruise. Do not use a hand-held device. Requires active Super Cruise plan or trial. Terms apply.