When the automotive industry discusses autonomous driving, attention often focuses on highways, urban navigation, and Robotaxis. Yet, the scenario that may first expose the gap in software capabilities might not be the highway at 100 km/h, but rather the parking lot where vehicles navigate at walking speed.
Varun Krishna Murthy, Consulting Manager at SBD Automotive, points out that parking is becoming a key proving ground for Software-Defined Vehicles (SDVs). The reason is straightforward: compared to open roads, parking lots feature lower speeds, relatively well-defined boundaries, and generally more controllable consequences in the event of an accident. Yet, they concentrate a host of the most intractable challenges in autonomous driving—narrow spaces, low-light environments, pedestrian crossings, pillar occlusions, faint parking lines, and obstacles of varying heights. The vehicle must simultaneously execute perception, localization, path planning, motion control, and human-machine interaction. If any single link proves unreliable, users will easily notice.
Therefore, while automated parking may appear to be merely an auxiliary feature, it is essentially a comprehensive examination of software-defined capabilities.
From "Turning the Wheel for the Driver" to "Handing Over the Task to the Car"
The logic of early automated parking required the driver to find a parking space, shift into reverse, and control the accelerator or brake, while the system solely handled steering. The subsequent introduction of remote parking allowed users to stand outside the vehicle and control its forward and backward movements via a key fob or smartphone. Taking it a step further is Automated Valet Parking (AVP): the driver exits the vehicle at the parking lot entrance, issues a command via a smartphone, and the vehicle autonomously searches for a space, parks itself, and returns to the pickup point when needed.
The distinctions among the three lie not only in their varying degrees of automation but also in the shift in responsibility and interaction dynamics. The first two features still center on the paradigm of "human supervising the machine," whereas true AVP requires the user to hand over the entire task to the vehicle.
Mercedes-Benz and Bosch previously operated an infrastructure-based AVP service at the P6 parking garage at Stuttgart Airport. The vehicle operates in coordination with cameras and control systems within the garage, allowing S-Class and EQS owners to exit in designated areas and let the vehicle park itself. In 2019, the two companies pioneered the world's first daily-use approval for SAE Level 4 automated parking at the Mercedes-Benz Museum parking lot in Stuttgart. In November 2022, the Airport P6 project received approval from the German Federal Motor Transport Authority (KBA), becoming the world's first commercial L4 automated parking service for mass-produced vehicles—a significant regulatory and engineering milestone. However, the service has since remained confined to the P6 parking garage and has not been scaled or replicated to other locations.
This also exposes the limitations of such a pathway: if the capability is highly dependent on parking garage retrofits, the speed of deployment hinges on the installation of cameras, communication networks, control centers, and operational platforms. What consumers purchase is not a universally applicable vehicle feature, but a service restricted to a few designated locations. Achieving L4 technically does not equate to achieving commercial scale.
Why Chinese Automakers Are Starting with Parking Lots
Chinese automakers have chosen a different route: relying as much as possible on vehicle-mounted sensors, computing platforms, and software to transform parking into a vehicle capability that can be continuously upgraded, gradually expanding its operational boundaries.
Brands such as XPeng, NIO, BYD, AITO, and Zeekr have expanded automated parking from a single maneuver to a complete workflow. The vehicle must not only reverse into a standard parking space but also understand parking lot layouts, memorize frequently used routes, navigate across floors to find target areas, and even depart from a designated space to seamlessly transition into assisted driving on campus or city roads.
For instance, NIO's parking space memory and remote parking features enable the vehicle to navigate to frequently used spaces and allow users to execute parking-in or parking-out maneuvers via their smartphones after exiting the vehicle. The official manual simultaneously emphasizes that, at this stage, users must still supervise the vehicle within a specified distance and remain ready to interrupt the operation at any time—illustrating that functional advancement and the boundaries of responsibility do not always progress in sync.
BYD's Denza, on the other hand, approaches the challenge from the chassis. Its rear-wheel dual-motor independent steering technology enables small-radius turns, while its "intelligent crab walk" allows large vehicles to move diagonally. Here, the software does more than merely identify parking spaces and calculate trajectories; it must deeply coordinate with the steering, braking, and powertrain systems. Zeekr's parking features, developed for ultra-narrow spaces, unmarked areas, and mechanical parking spots, reflect this same trend: the competition in intelligence is shifting from "whether automated parking exists" to "how many troublesome real-world parking scenarios can be handled."
This is precisely the core of Software-Defined Vehicles (SDVs)—hardware largely defines the boundaries of capability, while software determines how much of that capability users can actually invoke on a daily basis.
Why Parking Is More Suited Than Highways for Forming a Data Flywheel
The importance of highway assisted driving is unquestionable, yet variations in user experience are often constrained by regulations, map coverage, and road conditions. Parking, conversely, occurs almost daily, and its success or failure is relatively easy to measure: whether a parking space is recognized, how many steering adjustments were required, whether the vehicle is parked centrally, whether the system aborted due to an obstacle, and whether the user ultimately had to take over.
Subject to regulatory compliance, this data can be continuously uploaded to identify long-tail scenarios, train models, and optimize trajectory planning, which is then deployed to vehicles via Over-The-Air (OTA) updates. A parking failure caused by insufficient lighting could become a training sample for the next iteration of the perception algorithm; a specific pillar layout that frequently causes the vehicle to abort can drive adjustments in path planning strategies.
The resulting closed loop of "usage—feedback—training—upgrade" is precisely one of the key characteristics that distinguish SDVs from traditional vehicles. Traditional parking assistance systems are essentially finalized upon delivery; a software-defined parking system, however, can continuously add supported parking space types, expand operational boundaries, and reduce the frequency of user takeovers long after the vehicle is sold.
Parking also serves as a touchstone for evaluating the Electrical and Electronic (E/E) architecture. The data generated by cameras, millimeter-wave radars, LiDARs, and ultrasonic sensors must be fused by a central computing platform, and algorithmic decisions must be rapidly transmitted to the steering, braking, and powertrain systems. If an automaker struggles to consistently achieve cross-domain coordination and OTA iterations even in low-speed scenarios, its software capabilities on higher-speed, more open roads will inevitably be called into question.
Highway-First vs. Parking-First: Two Product Logics
Western luxury brands have historically prioritized highway scenarios, as long-distance driving and highway regulations align better with their traditional market structures. Chinese automakers, conversely, operate in high-density cities with crowded underground garages and a large demographic of novice drivers. Parking pain points are far more frequent and more easily translated into tangible value perceived by consumers.
It must be noted that this does not imply Western brands neglect parking technology—Mercedes-Benz and Bosch are, in fact, pioneers of L4 AVP. It is simply that their infrastructure-dependent pathway has limited the speed of scaling. The divergence lies more in their technological approaches and commercialization priorities: the former (Western luxury brands) tends to first prove the safety levels of autonomous driving, while the latter (Chinese automakers) prioritizes embedding software into high-frequency daily life scenarios; the former emphasizes technical certification, whereas the latter focuses on usage frequency, feature coverage, and OTA speed.
The supply chain is evolving accordingly. Companies such as Bosch, Valeo, AUMOVIO, and Aptiv are all advancing end-to-end parking solutions. However, for automakers to achieve differentiation, they cannot merely procure a fixed set of features; they must also master the data closed loop, algorithm updates, and whole-vehicle control interfaces. In the future, suppliers may no longer sell an "automated parking module," but rather a continuously iterating software stack, along with data tools and a validation framework.
Market forecasts vary significantly due to differing statistical methodologies and projection horizons. Furthermore, the subjects analyzed by these institutions primarily encompass automated parking systems (covering both vehicle-side features and infrastructure-side equipment), which cannot be directly equated to the in-vehicle parking software market alone, though the growth trajectory is consistent. Grand View Research projects that the global automated parking system market will expand from approximately $3.1 billion in 2026 to $10.7 billion by 2033, representing a compound annual growth rate (CAGR) of roughly 19.4%. Mordor Intelligence forecasts the market to grow from $3.36 billion in 2026 to $6.88 billion by 2031, representing a CAGR of about 15.4%, noting that software platforms are growing faster than the overall market.
The True Decisive Factor Remains Trust
The challenges of automated parking will not vanish as features multiply. Mobile connection drops, sensor obstructions or damage, the recognition of parking locks and low-lying obstacles, localization in underground garages, accident liability determination, and varying national regulatory requirements can all impede large-scale adoption. A single minor scratch could lead a user to disable the feature long-term. Therefore, beyond mere success rates, the system must ensure predictable behavior, clearly defined operational boundaries, and safe fallback mechanisms upon failure.
Ultimately, the significance of parking automation lies not in saving the driver a few turns of the steering wheel, but in establishing a new relationship of trust: whether the user is willing to step out of the driver's seat and entrust the vehicle to the software.
Highway autonomous driving showcases technological ambition, whereas parking tests whether software can truly integrate into daily life. The competition among Software-Defined Vehicles may not begin with a spectacular driverless revolution, but will quietly be decided in the final few dozen meters of the journey home each day. As Murthy noted, the most meaningful competitive advantage in autonomous driving may well be forged precisely at "walking speed."