Cruise outlines his plan for making robot axes a reality



The group of engineers that spoke Thursday night during an in-depth look at Cruise’s autonomous vehicle technology never mentioned Tesla’s name. They didn’t have to, even though the message was clear enough.

GM’s self-driving subsidiary Cruise presented a technical and deployment roadmap – at a granular level – designed to show how it has built autonomous vehicles that are safer and more scalable than any human-operated vehicle, including those equipped with advanced driver assistance systems are.

While Cruise was clearly advocating his own technology (not to mention recruiting new talent), the event also made a case for autonomous vehicles in general. Each engineer or product manager who spoke on Thursday presented various components, from using simulations to developing your own chips and other hardware to designing your app and the vehicle itself.

The branded “Under the Hood” event is based on comments from CEO Dan Ammann last month during GM’s Investor Day when he outlined the company’s plan to launch a commercial robotic taxi and delivery service starting with retrofitted Chevy Bolts and eventually, over the next several years, scale to an army of tens of thousands of purpose-built Origin AVs on the road.

cruise just licensed in California to provide commercial delivery services and is pending approval from charging fees driverless ride-hailing. However, Cruise believes that costs can be reduced enough to allow for rapid scaling.

Here is how.

Scale with simulations, not just verify the system

Cruise relies not only on simulations to prove its proof of safety, but also to scale to new cities without running millions of kilometers of testing beforehand.

The company must continue to map the cities it enters. However, it does not have to reassign cities in order to keep track of inevitable environmental changes such as lane changes or road closures. When Cruise goes to new cities, it starts with a technology called WorldGen, which enables accurate, large-scale generation of entire cities, “from their quirky layouts to the smallest details,” which enables engineers to test new operational domains, so Sid Gandhi, Head of Simulation Engineering Strategy at Cruise. In other words, WorldGen becomes the stage for future simulations.

To ensure the best possible world design, Cruise takes into account the lighting at 24 different times of the day and weather conditions, for example, and even goes so far as to systematically measure the light from a number of street lamps in San Francisco.

“If we combine a high-fidelity environment with a procedurally generated city, we can efficiently expand our business to new cities,” said Gandhi.

He then presented the technology for the “Road to Sim”, which transforms real-world events collected by AVs on the road into editable simulation scenarios. This ensures that the AV does not degrade through tests with already known scenarios.

“The Road to Sim combines perceptual information with heuristics learned from millions of miles in the real world to recreate a complete simulation environment from road data,” said Gandhi. “Once we have the simulation, we can actually create permutations of the event and change attributes like vehicle and pedestrian types. It’s a super simple and extremely powerful way to create test suites that accelerate AV development. “

Morpheus is available for certain scenarios that Cruise could not collect under real road conditions. Morpheus is a system that can generate simulations based on specific locations on the map. It uses machine learning to automatically fill in as many parameters as it wants to generate thousands of interesting and rare scenarios that will be used to test the AV.

“As we work on solving the long tail, we rely less and less on real testing because when an event rarely happens, it takes thousands of miles of road to test it properly and it’s just not scalable,” said Gandhi. “That is why we are developing a technology to explore large-scale parameter spaces in a scalable manner in order to generate test scenarios.”

The test scenarios also include the simulation of the reaction of other road users to the AV. Cruise’s system for this is called Non-Player Character (NPC) AI, 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 behavior.

“Morpheus, Road to Sim and NPC AI are working together in this really thoughtful way so that we can run more robust tests around rare and difficult events,” said Gandhi. “And it really gives us confidence that we can solve rare problems now and in the future.”

Generating synthetic data helps the Cruise AV track specific use cases, Gandhi said, referring specifically to identifying and interacting with emergency vehicles, presumably for no other reason than digging at Tesla, whose autopilot ADAS system has come under state control is repeated collisions with emergency vehicles.

“Emergency vehicles are rare compared to other types of vehicle, but we have to recognize them with an extremely high degree of accuracy. So we’re using our data generation pipeline to create millions of simulation images of ambulances, fire engines and police cars, ”said Gandhi. “In our experience, targeted synthetic data is about 180 times faster than collecting street data and millions of dollars cheaper. And with the right mix of synthetic and real-world data, we can increase relevant data in our datasets by an order of magnitude or more.” . “

Two customer-specific silicon chips developed in-house

During GM’s Investor Day in October, Cruise CEO Dan Ammann outlined the company’s plan to invest heavily in Origin’s computing power to cut costs by 90% over the next four generations so it can scale profitably. At the time, Ammann Cruises mentioned its intention to manufacture custom silicon in-house to reduce costs, but did not admit to using that silicon directly to build a chip – but did TechCrunch had its theories. On Thursday, Rajat Basu, chief engineer of the Origin program, confirmed these theories.

“Our fourth generation computing platform will be based on our in-house custom silicon development,” said Basu. “This was specially developed for our application. It enables focus and improves processing capacity, while unit costs and power consumption are significantly reduced. Computing is a safety-critical system and has built-in redundancy. In addition, there is an AV system that can process up to At 10 gigabit data per second, we end up using quite a lot of electricity. Our MLH chip allows us to run our complex machine learning pipelines in a much more focused manner, which in turn helps us to have more energy efficiently without compromising on the Performance. “

Cruise’s AI team has developed two chips: The sensor processing chip will handle the edge processing for the range of sensors such as cameras, radar and acoustics. The second chip, designed as a dedicated neural network processor, supports and accelerates machine learning applications such as the large multitask models developed by the AI ​​team. Basu says the Machine Learning Accelerator (MLA) chip is just the right size to solve exactly one class of neural networks and ML applications, and nothing more.

“This keeps performance at an extremely high level and ensures that we don’t waste energy on things that do not add value to us,” said Basu. “It can be paired with several external hosts or operated independently. It supports individual Ethernet networks up to 25 G with a total bandwidth of 400 G. The MLA chip that we are bringing into mass production is just the beginning. Over time, we’ll keep doing this even more while reducing power consumption. “

The cruise ecosystem

Cruise made it clear during its event that it was thinking not just about the AV technology needed to scale successfully, but the entire ecosystem that includes things like remote assistance operators to make the AV’s decision to confirm when unknown scenarios arise, customer service, a vehicle that people actually want to drive around in, and an app that can handle things like customer support and incident response efficiently and easily.

“To truly break the chasm of research and development into a popular product, it takes more than just artificial intelligence and robotics,” said Oliver Cameron, Vice President of Product at Cruise. “A safe, self-driving vehicle is not enough and is only the first step on a long, long journey. To truly develop and scale a competitive product that millions will adopt into their daily lives, you need to develop a multitude of different features and tools on a safe, self-driving foundation. How to implement these features is not obvious, especially if your company is still solving security problems. “

Similar video:



Leave A Reply