Autonomous Valet Parking Not Yet Ready for Implementation

Autonomous Valet Parking (AVP), an SAE L4 fully automated functionality that allows vehicles to park and be summoned autonomously without the presence of a driver, is available today in the Mercedes S-Class in Stuttgart, Germany, and Human Horizon's HiPhi X in China. These solutions use an infrastructure-based approach where all the intelligence is hosted in the parking garage. A computer in the facility receives information from several sources to decide what steering throttle and brake commands to send to each vehicle. The vehicle is drive by wire, in the same way as a remote-controlled car, and makes no decisions.

For effective interaction with the infrastructure sensors, vehicles must feature automatic transmission, electric parking brake, electric steering assistance, and an intelligent communication unit (e.g., cellular module, UWB, or V2X). As these components are commonly found in premium and mid-range vehicles today, the solution can be easily extended to support models from various manufacturers.

Several carmakers—including Mercedes, BMW, Cariad (A VW Group company), Ford, and Jaguar Land Rover—and Tier Ones– including Bosch, Kopernikus/Continental, and Valeo—showcased similar solutions at International Motor Show (IAA) Mobility 2021. However, despite their claims that AVP will soon be part of everyday life, ABI Research does not expect this infrastructure-based approach to scale in the B2C market.

The infrastructure-based AVP solutions available today sit within a highly constrained environment (e.g., three parking spots at the Stuttgart airport) and are, reportedly, rarely booked. Although another car park supporting AVP is soon to be inaugurated in London, according to ABI Research interviews with vendors in the parking infrastructure space, the list of car parks that will roll out infrastructure-based AVP in the following years is not impressive.

Despite being easy to deploy from a vehicle perspective, AVP will hardly scale in the Buisness to Consumer (B2C) market due to the extremely high installation and maintenance costs per facility. Besides, so far, no business model justifies the investments by the parts involved. Like we reported in our blog post Automated Parking: Valet Done from Your Phone, the parking industry—already suffering from tight margins—shows no interest in paying for the smart infrastructure costs. Car park operators rarely own the buildings where they operate and often have a five-year contract. That means they would need Return on Investment (ROI) within two to three years. Today, the current surcharge per reservation at the Stuttgart Airport is €10.00. Considering only three parking spaces were made available for AVP in the Stuttgart airport, and initial deployments are likely to be similar, it is insufficient to cover the infrastructure costs in the hundreds of thousands. Thus, ROI within the required timeline is unlikely. Tier Ones, such as Bosch, may be tempted to assume the costs initially, but they will need ROI at some point, which seems unlikely to happen. Carmakers and consumers are also unlikely to pay for the infrastructure costs.

Many industry players agree that infrastructure-based AVP makes for a solid Business-to-Business (B2B) business case in car manufacturing plants to move newly produced vehicles from one production point to the next, replacing actual drivers and reducing risks of damage. It justifies ROI because OEMs can save on salaries from drivers hired to do this job today, combined with a lower risk of vehicle damage. The B2B potential is as such that Continental has recently invested in Kopernikus Automotive, a Leipzig-based startup developing solutions for infrastructure-based maneuvering of vehicles, to create a cost-effective smart AVP infrastructure targeted at vehicle factories and logistics centers. Nevertheless, AVP requires a common technology standard to scale, and the demos featured at the IAA Mobility Show suggest that the industry is not close to reaching that point yet.

There is a second approach, called hybrid, where intelligence is situated partly in the car and partly in the infrastructure. However, it is deemed the most complex to deploy because of the strong dependency between carmakers and infrastructure players in architecting platforms and deploying technologies, the lack of clear definitions concerning what intelligence should be in the vehicle and the infrastructure, and, again, unclear ROI prospects.

ABI Research expects the AVP deployment based on vehicle intelligence to be the most solid path to market in the B2B vertical because of the lower infrastructure requirements. Nevertheless, challenges with this approach remain, including the vehicle's ability to provide accurate environment perception, prohibiting sensor and computing costs, and highly limited Operational Design Domain (ODD). Besides, minimum infrastructure requirements still must be in place. That includes drop-off/pick-up zones and the ability to stop all maneuvers in emergencies.

Lastly, accurate localization, which is crucial for L4 AVP, is tricky in indoor car parks. GPS signals fail to penetrate those locations with sufficient signal strength offering poor position resolution, and dead reckoning techniques are more challenging to apply in low-speed maneuvers. The leeway tolerance is much lower than in higher-speed scenarios. For instance, while a ten centimeter margin of error is perfectly acceptable in highway driving, it could cause a collision in a parking maneuver. Also, wheel sensors cannot easily distinguish a forward from a rear movement.

To this end, the International Organization for Standardization (ISO) is preparing an international standard, ISO 23374, to address localization issues and determine AVP standards. So far, the team drafting it has agreed on the requirement for Artificial Landmarks in the parking garage, although they recognize that it may generate prohibitive costs, preventing scalability. Thus, they are currently working on alternatives such as HD maps of parking facilities provided by parking service providers (e.g., Parkopedia) that require minimal investments from infrastructure players.

Due to the challenges in providing accurate sensing and localization based solely on the vehicle sensors and undefined minimum infrastructure requirements, AVP will remain subdued in the B2C segment until the end of this decade.