It’s a remarkably sunny day for San Francisco. The fog that often envelops the city in the morning has cleared, offering spectacular views of the Golden Gate and the bridge of the same name. I share a 2026 Mercedes-Benz CLA-Class with two other people, and as we enjoy the scenery, the car stops at a stop sign, checks that the intersection is clear, and turns left without human intervention.
This is Mercedes’ latest driver assistance system in action. It’s called MB.Drive Assist Pro and it’s coming to the US later this year in the redesigned CLA-Class. According to Mercedes, it corresponds to Level 2 on the Society of Automotive Engineers (SAE) autonomy scale, meaning drivers must remain alert and engaged at all times. But at the moment the car doesn’t seem to need the driver.
We’ve already seen Level 2 systems – General Motors’ Super Cruise, for example – and Mercedes has already achieved a higher level of automation with its Level 3 Drive Pilot system. But instead of pushing for greater automation for its own sake, Mercedes is refining Level 2 technology with Drive Assist Pro – making it more practical.
It’s as smart as it needs to be
Rather than just releasing a system to keep up with the automated driving zeitgeist, Mercedes is positioning Drive Assist Pro to make customers more likely to use it. This new system works on both motorways and country roads, but only if a preset route from the navigation system is followed. This makes it more likely that Drive Assist Pro will be available throughout the entire journey.
Drive Assist Pro also makes its U.S. debut (it launched in China in 2025) in an entry-level model. At press time, Mercedes had not yet confirmed the price of the system, but the CLA itself starts at $48,500 – not bad for a luxury brand electric vehicle. This is made possible by an extensive but not excessive hardware set including ten cameras, five radars and twelve ultrasonic sensors, but without the more expensive lidar units used in Mercedes’ Drive Pilot Level 3 system.
As a Level 2 system, Drive Assist Pro requires the driver to keep their eyes on the road, which is achieved with a driver monitoring camera on top of the infotainment touchscreen. However, rather than expecting drivers to simply watch until they have to take over, Mercedes took a more flexible approach that it calls “collaborative steering.” You can keep your hands off the wheel if you want, or take control if you don’t like how the system works. In any case, Drive Assist Pro remains active, so there are no ambiguities. Deactivation is done by applying the brakes or pressing the cruise control “cancel” button, just like Mercedes’ adaptive cruise control systems.
To deal with the chaotic nature of crowded city streets like those in San Francisco, Mercedes and software partner Nvidia emphasized a flexible decision-making process enabled by an AI that is “like an LLM but for (computer) vision,” Ali Kani, Nvidia’s vice president of automotive, told Daily Sparkz and other media outlets ahead of this test drive. The model helps the system synthesize decisions from its sensor data and is supported by a conventional software “safety stack” that ensures the system always follows traffic rules.
It drives like a human
All of this, in practice, allowed the Mercedes CLA to navigate the streets of San Francisco, as well as the Waymo and Zoox robotaxis with which it shared those streets – just without the driver having to climb into the back seat and take a nap.
Drive Assist Pro accelerated and braked smoothly while using the CLA’s forward-facing cameras to read traffic lights and signs to know when to stop or when not to turn right on red. The system is programmed with such region-specific rules and will always follow them regardless of the situation. For example, you wouldn’t cross a solid yellow line to avoid a car parked in the double alley. Drivers can still override this behavior at their own discretion; After making sure the path was clear, the driver took over the steering to avoid the obstacle.
Conversely, when we encountered two double-parked cars on a road with broken yellow lines, the system tried to avoid both at once. But the driver didn’t like the distance to an oncoming car and pushed the CLA back into the lane. It’s this ability to keep the driver informed at all times that turns Drive Assist Pro’s Level 2 automation from a limitation into an advantage.
Further proof that the transition to manual control was seamless when the car did hesitate in a complex situation. In fact, it wasn’t a surrender at all; The driver simply saw a problem and intervened. In this case, it involved a left turn through an intersection with crossing pedestrian traffic blocked by a gated intersection and another car attempting to turn right into the same lane as us. Human intuition was able to analyze this situation more easily.
It is a more realistic approach to automated driving
As we drove through the city, the numerous Waymo Jaguar I-Pace SUVs, their exteriors peppered with whirring sensors, reminded us of the difficulties of expanding autonomous driving technology. A driverless experience is Silicon Valley’s dream, but after more than a decade in development, it’s just barely accessible to the public. Even issues such as the recent unexpected shutdown of Waymo vehicles during a power outage are still emerging, and the business case for robotaxi services is not yet proven.
The collaborative approach between driver and computer developed by Mercedes shows a more realistic path for this technology in the near term – albeit in a more limited capacity. There are other Level 2 systems, including Super Cruise and Teslas with the deceptively named “Full Self-Driving.” But Mercedes has put more thought into how the interfaces will need to change if these systems are to operate further away from the more predictable highway environment and offer drivers more opportunities to use the technology.
MB.Drive Assist Pro appears to be the next evolutionary step for driver assistance systems, not because it aims to make the leap towards fully automated driving, but because it makes optimal use of the sensor and computer technology available today. It’s a product, not a promise.
