Why Building a Smart Driving System Feels Like a Very Slow Hobby

The Long Game of Platform Building

There's a certain kind of hobby that doesn't reward you quickly. You spend months, sometimes years, on something that barely seems to move. Then one day, you look back and realize you've built something enormous. Volkswagen's approach to smart driving feels exactly like that.

In 2012, the third-generation Audi A3 became the first car built on Volkswagen's MQB platform. A few months later, the seventh-generation Golf rolled off the same modular system. MQB wasn't a fixed chassis—it standardized the engine position, the relationship between pedals and front axle, but allowed different wheelbases, widths, and tuning. That's how the Golf, Passat, and Tiguan could share parts and manufacturing while still being distinctly different cars. Ten years later, MQB cars had surpassed 32 million units. That's the kind of patience most of us can only dream of.

Treating Driving Like a Craft

Now Volkswagen is applying that same patient, platform-first thinking to something less tangible: how a car drives. It's not about a specific model or a single feature. It's about building a system that can be reused, refined, and scaled across different vehicles, just like a hobbyist perfecting a technique over decades.

For years, Volkswagen's software arm, CARIAD, struggled to unify its smart driving efforts. The pieces were scattered—different teams working on assisted driving, cockpit, and electronic architecture, never quite converging into a single platform. It wasn't until recently that Volkswagen China pulled the threads together. By the end of 2025, a highway navigation system based on the CEA architecture landed on the 与众 07 and 06 models. Then, in July, the group deepened its partnership with Horizon Robotics, folding AI foundation models and L3/L4 research into the same pipeline. Starting in the third quarter, a full-scenario HS8 system began rolling out across seven models from three joint ventures.

The Training Ground: GAIA 2.0

If building a smart driving system is a hobby, then GAIA 2.0 is the workshop. It's a data and simulation engine that generates edge cases—like sudden cut-ins, pedestrians crossing, or temporary construction zones—that are rare in the real world but critical to master. In the simulation, it can replay these scenarios thousands of times, tweaking weather, lighting, and the behavior of other road users. It can even take a single front-view image and generate surround-view footage, or create lidar point clouds from visual data, so the same dataset can serve different sensor configurations.

This is the kind of meticulous practice that separates a casual dabbler from a true enthusiast. You don't just wait for real-world encounters; you manufacture them, again and again, until the response becomes second nature.

From Data to Driver: The HS8 System

The data from GAIA 2.0 feeds into a foundation model called HS (Hyper Sense). After distillation and quantization, it becomes HS8—a compact model that runs on production car chips. HS8 uses a one-stage end-to-end model to generate driving decisions, cutting down the multiple conversions between perception, prediction, and planning. But it still keeps safety mechanisms, compliance checks, and post-processing intact.

Depending on the car, HS8 can take different forms. The pure-vision version uses a single Horizon Journey 6M chip, 11 cameras, and 128 TOPS of computing power. It's first installed in the ID. ERA 5S and will later appear in the Jetta M6. The higher-end version uses the Journey 6H chip, a lidar, and 420 TOPS. Volkswagen develops the tools, models, and safety standards centrally, while hardware, chassis tuning, and human-machine interface vary by model.

The Choreography of Hardware and Software

The CEA (Central Electronic Architecture) is what ties it all together. Developed in China, it lets the smart driving system communicate with steering, braking, chassis, cockpit, and HMI through a unified interface. It works across pure electric, plug-in hybrid, range-extender, and even gasoline cars. Software updates are decoupled from vehicle generations, with a target of quarterly OTA pushes.

Think of it as a driving school. GAIA 2.0 is the training ground that can change weather and road conditions at will. HS is the curriculum. HS8 is the trained driver, compressed into the car's brain. And CEA is the hands and feet, letting that driver control any car's steering, brakes, and chassis through a common set of commands. This whole system can be tailored to different models, much like a hobbyist adapting a core skill to different projects.

Why Start with the Entry-Level Cars?

Volkswagen's decision to roll out HS8 first on lower-priced models might seem odd. But as CARIAD CEO Han Hongming puts it, "Starting from entry-level models and moving up gives us more data." The more cars you sell, the more real-world driving scenarios you encounter, and the smarter the model becomes. Saving costs is just the first benefit; data is the real payoff. That's why Volkswagen is pushing smart driving into cars under 150,000 yuan—the volume sellers are the data sources.

A Polite, Confident Driver

HS8 is designed to handle city, highway, and parking scenarios. It can distinguish between multiple lane signals and right-turn-only lights, using signal states, navigation, and surrounding traffic to make decisions at intersections, roundabouts, and when facing slow vehicles or pedestrians. It's trained on data from about 100 professional drivers with over 20 years of experience—the kind who drive fast but safely, not the reckless sort.

The system doesn't just mimic human driving results; it uses "residual learning" to understand the gap between the vehicle's trajectory and an optimal one, then calibrates with the chassis actuators to turn model decisions into smooth steering, braking, and body movements. In plain terms, that means braking to a stop without jerking, following in traffic without constant hard stops, and braking linearly when a pedestrian crosses while maintaining a safe distance. These are the hallmarks of a skilled driver, and they're what makes the experience feel refined.

The First Test: ID. ERA 5S

The ID. ERA 5S is the first car to put HS8 to the test. It's a plug-in hybrid aimed at mainstream families, with a pre-sale price between 115,900 and 145,900 yuan. The Max version includes city navigation assist and costs 145,900 yuan. Its CLTC fuel consumption is just 2.82 liters per 100 km when the battery is depleted, with a combined range over 2,000 km. That's a perfect platform to gather real-world data across long distances and varied conditions.

For Volkswagen, the ID. ERA 5S is just the first graduate of this "driving school." The real validation will come when the same underlying data, models, interfaces, and safety standards are applied across seven models from three joint ventures. Can they maintain consistent capability while allowing each car to keep its own hardware, chassis tuning, and interaction style? That's the question.

In the fossil-fuel era, Volkswagen's platform defined which mechanical parts had to be identical. In the smart car era, that scope extends to data and models. With HS8, Volkswagen is doing what it's always been best at: making the common parts a platform, and leaving the different parts to each car. It's a slow, methodical process—much like a hobby that demands years of practice before you see the payoff. But when it works, the results can be staggering.