The series of engineers who spoke Thursday night during a deep dive in Cruise’s autonomous vehicle technology never mentioned by Tesla Last name. It wasn’t necessary, even if the message was clear enough.
GM Autonomous driving subsidiary Cruise unveiled a technical and implementation roadmap – at a granular level – that aimed to show how it was built autonomous vehicles safer and more scalable than any human-driven vehicle, including those equipped with advanced driver assistance systems.
While Cruise was clearly supporting its own technology (not to mention trying to recruit new talent), the event was also a topic for autonomous vehicles in general. Each engineer or product manager who spoke on Thursday presented various components, from how they use simulations and development of their chips and other hardware to the design of their app and vehicle itself.
The “Under the Hood” event with the brand based on the comments of the CEO Dan Ammann made last month during GM’s investor day, where he exposed the company’s plan to launch a commercial robotaxi and delivery service starting with retrofitted Chevy Bolts and eventually scale up to an army of tens of thousands of purpose-built AV Origin on the road over the next few years.
Cruise just got approval in California to perform commercial delivery services, and is still a permit away from being able to charge driverless call. However, Cruise thinks it will be able to cut costs enough to go up and down quickly.
Using simulations to scale, not just test the system
Cruise relies on simulations not only to prove he is out of his safety case, but also to scale new cities without having to run millions of miles of testing in them first.
The company will still have to map the cities it enters. But it won’t have to remap cities to track environmental changes that inevitably occur, such as lane changes or road closures. When Cruise goes to new cities, he starts with a technology he calls WorldGen, which is said to achieve an accurate, large-scale generation of entire cities, “from their wacky layouts to the smallest details,” which allows engineers to test new operational designs. domains, according to Sid Gandhi, head of simulation technical strategy at Cruise. In other words, WorldGen becomes the stage where future simulations are set up.
To ensure optimal creation of the world, Cruise takes into account elements such as lighting in 24 different hours of the day and unique weather conditions, even systematically measuring the light from a series of street lamps in San Francisco.
“When we combine a high-fidelity environment with a procedurally generated city, that’s when we unlock the ability to efficiently extend our business to new cities,” Gandhi said.
He then developed the technology for “Road to Sim,” which transforms real events that have been collected by the AVs on the road into editable simulation scenarios. This ensures that the AV does not regress by testing the scenarios it has already seen.
“The Road to Sim combines perception information with heuristics learned from our millions of real world miles to recreate a complete simulation environment from road data,” Gandhi said. “Once we have the simulation, we can actually create event permutations and modify attributes such as vehicle and pedestrian types. It’s a very simple and extremely powerful way to create test suites that accelerate AV development.”
For specific scenarios that Cruise was unable to pick up in real-world road conditions, there’s Morpheus. Morpheus is a system capable of generating simulations based on specific locations on the map. It uses machine learning to automatically enter as many parameters as you want to generate thousands of interesting and rare scenarios against which the AV tests.
“As we work to fix the longtail, we will rely less and less on real world testing because when an event occurs that rarely occurs, it takes thousands of miles to test it properly and it is not scalable,” Gandhi said. “So we are developing technology to scale-scale parameter spaces on a large scale to generate test scenarios.”
The test scenarios also include simulating how other road users react to AV. Cruise’s system for this is called non-player character AI (NPC), which is usually a video game term, but in this context it refers to all the cars and pedestrians in a scene that represent complex multi-agent behaviors.
“So Morpheus, Road to Sim and NPC AI work together in this very thoughtful way to allow us to perform more robust tests on rare and difficult events,” Gandhi said. “And it really gives us the confidence that we can solve rare problems now and similar problems in the future.”
Synthetic data generation helps the Cruise AV target specific use cases, Gandhi said, specifically indicating identification and interaction with emergency vehicles, presumably for no other reason than to take a look at Tesla, whose ADAS Autopilot system has come under federal scrutiny for repeated accidents against emergency vehicles.
“Emergency vehicles are rare compared to other types of vehicles, but we need to detect them with extremely high accuracy, so we use our data generation pipeline to create millions of ambulance simulation images,” fire truck Other police cars, “said Gandhi.” In our experience, targeted summary data is about 180 times faster than road data collection and millions of dollars less. And with the right mix of synthetic and real data, we can increase the relevant data in our datasets by an order of magnitude or more. “
Two in-house developed custom silicon chips
During GM’s Investor Day in October, Cruise CEO Dan Ammann outlined the company’s plan to invest heavily in Origin’s computing power in order to cut costs by 90% over the next four generations so it can scale profitably. At the time, Ammann mentioned Cruise’s intention to produce custom silicon in-house to reduce costs, but did not openly admit to using that silicon to build a chip, but TechCrunch had its theories. On Thursday, Rajat Basu, chief engineer of the Origin program, validated these theories.
“Our fourth generation processing platform will be based on our in-house custom silicon development,” said Basu. “This is built specifically for our application. It enables focus and improves processing capacity, significantly reducing part costs and power consumption. Computing is a safety-critical system and has built-in redundancy. Add to this an AV system that processes up to 10 gigabits of data per second, we end up consuming a fair amount of power. Our MLH chip allows us to run our complex machine learning pipelines in a much more targeted manner, which a in turn helps us be more energy efficient without compromising performance. “
Cruise’s AI team has developed two chips: the sensor processing chip will handle edge processing for the range of sensors such as cameras, radar and acoustics. The second chip, designed to be a dedicated neural network processor, supports and accelerates machine learning applications such as those large multitasking models developed by the AI team. Basu says the machine learning accelerator (MLA) chip is just the right size to solve exactly a certain class of neural networks and ML applications, and nothing more.
“This keeps performance at an extremely high level and ensures that we are not wasting energy on doing something that is not adding value to us,” said Basu. “It can be paired with multiple external hosts or run standalone. It supports single Ethernet networks up to 25G with a total bandwidth of 400G. The MLA chip we’re putting into mass production is just the beginning. Over time. we will continue to make this even more efficient while reducing energy consumption. “
The ecosystem of the cruise
One thing Cruise made clear at his event is that he has not only thought about the AV technology needed to grow successfully, but also the entire ecosystem, which includes things like remote assistance operators to validate the AV’s decision when runs into unfamiliar scenarios, customer service, a vehicle that people really want to get around in, and an app that can efficiently and easily handle things like customer support and incident response.
“To truly cross the gap from research and development to a beloved product requires more than just artificial intelligence and robotics, “said Oliver Cameron, Cruise’s vice president of product, at the event.” self-driving vehicle alone it is insufficient and simply the first step on a long, long journey. To truly create and scale a competitive product that is adopted by millions of people in their daily lives, it is necessary to create a set of differentiated features and tools on a safe autonomous driving basis. How these features need to be implemented isn’t obvious, especially if your business is still busy solving security issues. “