Mercedes-Benz brings autonomous driving to city streets with new AI-powered system

Mercedes teaches its driver assist how to handle surface streets

There's some debate as to when adaptive cruise control first showed up, but if you ask Mercedes-Benz, it will say in 1999 with that year's S-Class. Instead of just keeping a set speed, radar-enabled adaptive cruise control allowed the car to react to deceleration by the car ahead, and thus was created the first partially automated car. From there, automakers added a function to keep cars in their lanes, and now we have location-aware, GPS-geofenced vehicles that, as long as the driver is paying attention, will do most of the driving -- on the highway at least. But the goal for developers of both autonomous and partially automated vehicles is to remove as much of the burden of driving from the human as possible, not just on controlled access highways but at lower speeds, on surface streets. Which is what Mercedes' latest Drive Assist Pro has been designed to do. And after a recent demo -- albeit from the passenger seat -- on the streets of downtown San Francisco, it appears to be a very credible effort. CLA gets it first The big, powerful, comfortable S-Class is normally the standard-bearer for the latest and greatest tech Mercedes has cooked up, but not always. In December we drove the production version of its new entry-level EV, the CLA. At under $50,000, the sleek Mercedes sedan (or four-door coupé) is already available with the current version of the automaker's Drive Assist suite, with better control of braking and deceleration. A particular improvement, which I'm not sure made the final version of our first drive report, is the way you can use the brake while adaptive cruise control is active without canceling the system. Light applications of the brake -- to shed a few miles per hour, not to conduct an emergency stop -- will slow the car, which then resumes its original speed, much the same way you have always been able to apply the throttle to temporarily speed up while using cruise control. Drive Assist Pro takes this collaborative approach between the car and driver and runs with it. Given a destination to work with, it knows which lanes you'll need ahead of time, and the car reads both stop signs but also traffic lights. It even detects, and slows down for, speed bumps. On a 20 minute drive around the Waymo-clogged streets of the tech industry's favorite city, the engineer in the driving seat didn't have to intervene once, although I believe at least a couple of colleagues' demos got confused by human crosswalk attendants moving around with their stop signs. The CLA drove at safe and legal speeds, knew how to handle construction zones, and wasn't flummoxed by one of those most common of city driving annoyances, the double-parked car. The time the car takes to come to a complete stop at stop signs can and will irritate human drivers behind you; it's definitely not a California stop. Needs to be an SDV All of this is possible thanks to the CLA being what the industry calls a software-defined vehicle. Four powerful computers run all the electronics, rather than dozens and dozens of discrete black boxes. One of those computers is (of course) from Nvidia -- that company's Orin, which handles things like perception and path planning. "We completely elevated our autonomous driving stack. It is no longer on a rule-based stack," explained Magnus Östberg, chief software officer at Mercedes-Benz. Now it uses an end-to-end AI model, "which of course is giving you some basic advantages. When it comes to parking, for example, [it offers] much faster navigation of parking lots..., moving in and out of the parking lots, but also already you find... how it's on the highway and how it actually follows the lane and moving across it," Östberg said. As seamless as the experience appeared, the human behind the wheel remains the one in charge. I don't love the SAE levels as a scheme for explaining the spectrum of partially automated and autonomous driving these days, but Drive Assist Pro is what an engineer might call "level 2 ++," if "level 2+" is something tightly geofenced like Super Cruise and true level 3 is Mercedes Drive Pilot, which is only available in Nevada and California and only works in relatively low-speed freeway traffic jams. The closest thing might be Tesla's much-maligned driver assist, which has long had the goal of letting its users go from point to point with no human intervention. I shan't dwell on that company's record of either safety or success (although apparently, many years after Musk first promised a coast-to-coast drive, someone actually finally pulled it off in late 2025). Needless to say, Mercedes-Benz has a very different approach to safety. There are redundant sensor modalities, not just a smattering of cameras. The end-to-end AI model tokenizes input and output trajectories, and there's a safety guardrail with a rules-based model to make sure any AI trajectory mistake is caught before it's driven. There's certainly no "mad max" setting to make it break speed limits, unlike those camera-only Teslas. Mercedes has already launched Drive Assist Pro in China and told Ars the safety certification process is complete for the US, with it arriving late this year in the CLA and then other Mercedes vehicles as they get their SDV upgrades during midlife refreshes. As for Europe, that will require some regulation changes, we believe.

Mercedes to offer autonomous driving tech for US city streets

Jan 5 (Reuters) - Mercedes-Benz (MBGn.DE), opens new tab said on Monday it will launch a new advanced driver-assistance system in the United States later this year that lets its vehicles operate autonomously on city streets under driver supervision. The system, which enables a vehicle to drive from a parking lot to a destination, navigating city intersections, making turns and obeying traffic lights, is likely to pose competition to Tesla (TSLA.O), opens new tab, the only automaker that offers a similar product, called Full Self-Driving, in the United States. Mercedes' system, called MB.DRIVE ASSIST PRO, has been on sale in China since late last year. The system will cost $3,950 for three years in the United States. Customer can also choose on a monthly or yearly subscriptions, but the pricing for those will be disclosed later. Tesla's Full Self-Driving package costs about $8,000 as a one-time purchase or $99 per month as a subscription. Most automakers limit self-driving features in personal vehicles to highways, where traffic patterns are more predictable. Cities pose tougher challenges, including pedestrians, cyclists and unexpected situations. Tesla (TSLA.O), opens new tab is the only automaker that, with its Full Self-Driving system, allows self-driving on city streets. But like Tesla, Mercedes' system will require drivers to remain alert and ready to intervene at all times. Mercedes' push into urban driving assistance shows how software advances are moving autonomous technology from limited testing toward commercial rollout. Safety concerns and regulation still constrain full autonomy in personal vehicles. Tesla Chief Executive Elon Musk previously said he would flood city streets with autonomous vehicles that needed no human intervention. That has not happened yet. Instead, Tesla has focused on incremental improvements in FSD and has launched a small robotaxi service in Austin, Texas with safety monitors. Investors still view autonomous technology as a potential long-term revenue driver for automakers. Mercedes said the system uses about 30 sensors, including cameras, radar and ultrasonic sensors. Those sensors feed data to a computer that can process up to 508 trillion operations per second. Nvidia said the new Mercedes-Benz CLA, the brand's first vehicle featuring the MB.OS platform, will feature driver-assistance features powered by the chip designer's "DRIVE AV" software, AI infrastructure and accelerated compute. The system supports over-the-air updates for future improvements to the autonomous driving tech. Reporting by Akash Sriram in Bengaluru and Abhirup Roy in Las Vegas; Editing by Tasim Zahid Our Standards: The Thomson Reuters Trust Principles., opens new tab * Suggested Topics: * Autos & Transportation * ADAS, AV & Safety * Manufacturing * Sustainable & EV Supply Chain * EV Strategy Abhirup Roy Thomson Reuters Abhirup Roy is a U.S. autos correspondent based in San Francisco, covering Tesla and the wider electric and autonomous vehicle industry. He previously reported from India on global corporations, capital markets regulation, white-collar crime, and corporate litigation. Contact him at (415) 941-8665 or connect securely via Signal on abhiruproy.10 Akash Sriram Thomson Reuters Akash reports on technology companies in the United States, electric vehicle companies, and the space industry. His reporting usually appears in the Autos & Transportation and Technology sections. He has a postgraduate degree in Conflict, Development, and Security from the University of Leeds. Akash's interests include music, football (soccer), and Formula 1.

I Tried Mercedes' Answer To Tesla Full Self-Driving. It Handles City Streets -- With One Major Caveat

Riding shotgun next to Lucas Bolster in a prototype Mercedes-Benz CLA-Class, I tried to keep up as he explained his company's artificial intelligence strategy. I was there to see how MB.Drive Assist Pro -- Mercedes' new urban-focused automated driving system -- would handle San Francisco traffic, not just highways. Bolster, Mercedes' manager of automated and assisted driving, said the system runs two AI models in parallel. "That certainly helps with validation, and it helps us achieve our safety goals," he said. Interesting enough, but I was getting impatient. When would I see it in action? Then it hit me. MB.Drive Assist Pro had been on the whole time. It was so seamless that I didn't even realize it. Maybe that's because of its signature feature: It requires drivers to keep their hands on the steering wheel at all times. But if they're willing to do that, and keep their eyes on the road and be ready to intervene, then their next Mercedes-Benz could become a great "partner" for navigating the hazards of city driving. At the very least, this CLA seemed to fit in just fine amid San Francisco's endless sea of Waymo robotaxis. "All part of the fun, right?" Bolster said. Maybe, but this is still something very new and very different. Hands-on automated driving features -- Level 2 advanced driver-assistance systems (ADAS), in technical parlance -- have spread across the auto industry like wildfire in recent years. So long as a driver keeps their hands on the wheel and eyes on the road, the car can steer to stay in its lane on highways, slow and stop if a traffic jam arises, and even change lanes automatically. The next step is hands-free highway driving, now offered by Tesla, General Motors, Ford and others, where drivers only need to monitor the road and be ready to intervene. Beyond that is "eyes-off" Level 3 autonomy, which Mercedes technically already offers -- albeit in a highly limited form. All of these systems share a major constraint: where they work. Consumer ADAS has largely been confined to highways or, at best, low-speed traffic jams. Everything else has been left to robotaxis like Waymo. Tesla has historically been the lone exception, offering ADAS for urban environments. Its Full Self-Driving (Supervised) can operate hands-free in cities and promises to navigate all the way to a programmed destination. Though, as the feature's name suggests, drivers must remain ready to take over. Despite notable progress -- including a recent cross-country drive completed entirely with FSD -- the system has drawn criticism and regulatory scrutiny over how its capabilities are marketed. Now Mercedes is preparing to enter that space with MB.Drive Assist Pro. The system promises "intuitive urban point-to-point driving," capable of handling city traffic, intersections, roundabouts, parking maneuvers and, eventually, finding its own parking spot. On the CLA-Class, the system is powered by 10 cameras, five radar sensors, 12 ultrasonic sensors and a powerful onboard computer from Nvidia. It is trained on AI models using real-world data from Mercedes' cars on the road right now as well as simulated models, and it runs multiple redundant algorithms at once to make sure nothing was overlooked or misunderstood. The system will also get frequent over-the-air software updates to improve performance, perhaps quarterly, Mercedes officials said. The approach to hardware stands in contrast with other ADAS setups. Tesla, famously, relies solely on cameras and AI for FSD. And Rivian will soon explore urban point-to-point driving that will incorporate lidar as well. But Mercedes insists that this hardware, software and AI approach is one that can truly guarantee the safety of urban autonomy. "There's not a passenger car they can drive this safely, other than when Mercedes is showing today," said Ali Kani, who leads Nvidia's automotive platform. "We are doing something that the entire industry has been trying to do for 10 years, and no one has done it before." Like Tesla and others, Mercedes' use of AI means that MB.Drive Assist Pro does not have to be "trained" on every single route it drives -- unlike autonomous vehicles in the 2010s, which essentially had to relearn how to drive in every city they ventured into. It also means the system can learn over time, including what its own driver prefers. But a big part of ensuring safety means keeping your hands on the wheel -- and that may never change. "I think we don't see a good safety case for that in urban driving right now," Bolster said. "As soon as your hands are off the wheel, you automatically bring in a prolonged reaction time." Yet it seemed a bit counterintuitive to me. I'm generally a fan of Ford's BlueCruise and General Motors' Super Cruise. I don't like having to keep my hands on the wheel during long, freeway road trips if I don't have to. So if I have to keep my hands on the wheel in a city, why not just drive myself? For this, Bolster had an interesting answer: Ever drive in a brand new city for the first time? "It's stressful," he said. "You're trying to navigate and figure out which lane you need to be in to take which turn... It's a little bit more relaxing to supervise the system than it is to do it all yourself." I'm not sure I agree, since I find the emotional and mental energy needed to "supervise" FSD in a city to be more taxing than driving the old-fashioned way. But on our short loop through San Francisco evening traffic, the CLA performed admirably. Bolster kept his hands on the wheel and the car did the rest, navigating three- and four-way stops and other traffic (including of the non-human Waymo sort) with ease. You can also step in to steer yourself without fully disengaging the system, which Bolster calls "collaborative" steering -- something that also adapts to the user's driving style. "We want to give a more aggressive driver the opportunity to just step in if they need, if it's going too slowly, or if it's acting too cautiously," Bolster said. "You can always accelerate or stay with the system, and it doesn't penalize you for that." Our CLA got a little tripped up by a double-parked car, even if it can "read" another car's lights to see if they're parked or waiting to move forward. But in another instance, it predicted where another oncoming car was trying to go in a crowded intersection, and then drove off when it had a clear path. The CLA also yielded to pedestrians and e-bikes, and also navigated an unprotected left turn. Not bad at all. It's hard to render judgment without being behind the wheel myself. But Mercedes' approach to urban autonomy does seem promising, even in prototype form. A lot more people will be able to experience it starting this year. MB.Drive Assist Pro debuts on the CLA-Class, but will spread to more models soon. Its pricing has not been released yet, but it will be available at purchase or as a subscription feature. Considering that more sophisticated ADAS now ranks among the most wanted (and most subscribed-to) features on new cars, Mercedes is certainly banking on the idea that many drivers will want to try it out. But that demand is primarily for hands-free highway systems. Can Mercedes convince its buyers that this is an ideal way to contend with city traffic? "I think there's potential for a broad section of customers," Bolster said. "Our objective is that the system would be a value-add. It shouldn't just be for the person who wants to watch it do cool things. It should also bring comfort to the to the drive."

NVIDIA DRIVE AV Software Makes Production Debut in New Mercedes-Benz CLA | AIM

Mercedes-Benz's latest CLA recently earned a five-star safety rating from the European New Car Assessment Programme (Euro NCAP) NVIDIA's DRIVE AV software is set to hit US roads later this year with its first production deployment in the all-new Mercedes-Benz CLA, as per the NVIDIA's official blog. The launch brings enhanced Level 2, point-to-point driver-assistance capabilities to consumers and signals the start of broader adoption of NVIDIA's full-stack automotive software. The new CLA is Mercedes-Benz's first vehicle built on its MB.OS operating system and integrates NVIDIA DRIVE AV software, AI infrastructure, and accelerated compute. The system powers advanced driver-assistance features under the MB.DRIVE ASSIST portfolio, with the architecture designed to support over-the-air updates for future upgrades and new capabilities, available both ex-factory and through the Mercedes-Benz digital store. Mercedes-Benz's latest CLA recently earned a five-star safety rating from the European New Car Assessment Programme (Euro NCAP), with the performance of its active safety and accident-avoidance systems contributing to the top score. NVIDIA said the deployment underscores how AI-driven software and data-centric development are becoming central to vehicle safety and performance. Ali Kani, vice president of automotive at NVIDIA, said that as the automotive industry embraces physical AI, NVIDIA is the intelligence backbone that makes every vehicle programmable, updatable and perpetually improving through data and software At the core of the system is NVIDIA DRIVE AV's dual-stack architecture, which combines an end-to-end AI driving stack with a parallel, classical safety stack built on NVIDIA's Halos safety system. This approach adds redundancy and safety guardrails, allowing vehicles to learn from large volumes of real-world and synthetic driving data while operating within defined safety parameters. The unified architecture enables advanced Level 2 automated driving features, including point-to-point urban navigation, proactive collision avoidance, automated parking in tight spaces, and cooperative steering between the driver and the system. NVIDIA said its deep-learning models allow vehicles to interpret traffic holistically, respond to vulnerable road users such as pedestrians and cyclists, and assist drivers from one address to another in complex city environments. Beyond in-vehicle intelligence, NVIDIA and Mercedes-Benz are also applying AI to vehicle manufacturing. Using NVIDIA Omniverse and digital twin technology, engineers can design and optimise factory layouts and assembly lines virtually, reducing downtime and accelerating development cycles. Simulation platforms such as Omniverse and NVIDIA Cosmos also allow driving software to be tested and validated extensively in virtual environments before real-world deployment. NVIDIA's automotive strategy is built around a cloud-to-car development pipeline that spans AI training, simulation, and in-vehicle compute. Massive GPU-powered systems train driving models on global datasets, simulation tools convert real-world miles into billions of virtual test scenarios, and NVIDIA DRIVE AGX and Hyperion platforms handle real-time perception, sensor fusion, and decision-making inside the vehicle. NVIDIA said the Mercedes-Benz deployment is part of a broader effort to bring its full-stack software and AI infrastructure to automakers worldwide, enabling scalable integration of intelligent driving and safety features while simplifying future upgrades.

Mercedes to offer autonomous driving tech for US city streets

Jan 5 (Reuters) - Mercedes-Benz said on Monday it will launch a new advanced driver-assistance system in the United States later this year that lets its vehicles operate autonomously on city streets under driver supervision. The system, which enables a vehicle to drive from a parking lot to a destination, navigating city intersections, making turns and obeying traffic lights, is likely to pose competition to Tesla, the only automaker that offers a similar product, called Full Self-Driving, in the United States. Mercedes' system, called MB.DRIVE ASSIST PRO, has been on sale in China since late last year. The system will cost $3,950 for three years in the United States. Customer can also choose on a monthly or yearly subscriptions, but the pricing for those will be disclosed later. Tesla's Full Self-Driving package costs about $8,000 as a one-time purchase or $99 per month as a subscription. Most automakers limit self-driving features in personal vehicles to highways, where traffic patterns are more predictable. Cities pose tougher challenges, including pedestrians, cyclists and unexpected situations. Tesla is the only automaker that, with its Full Self-Driving system, allows self-driving on city streets. But like Tesla, Mercedes' system will require drivers to remain alert and ready to intervene at all times. Mercedes' push into urban driving assistance shows how software advances are moving autonomous technology from limited testing toward commercial rollout. Safety concerns and regulation still constrain full autonomy in personal vehicles. Tesla Chief Executive Elon Musk previously said he would flood city streets with autonomous vehicles that needed no human intervention. That has not happened yet. Instead, Tesla has focused on incremental improvements in FSD and has launched a small robotaxi service in Austin, Texas with safety monitors. Investors still view autonomous technology as a potential long-term revenue driver for automakers. Mercedes said the system uses about 30 sensors, including cameras, radar and ultrasonic sensors. Those sensors feed data to a computer that can process up to 508 trillion operations per second. Nvidia said the new Mercedes-Benz CLA, the brand's first vehicle featuring the MB.OS platform, will feature driver-assistance features powered by the chip designer's "DRIVE AV" software, AI infrastructure and accelerated compute. The system supports over-the-air updates for future improvements to the autonomous driving tech. (Reporting by Akash Sriram in Bengaluru and Abhirup Roy in Las Vegas; Editing by Tasim Zahid)