Over-The-Air upgrade: How to develop the concept successfully

 NASA used a precursor to these technologies to bring the Apollo 13 astronauts back to Earth. Lockheed Martin claims these types of solutions are one of six game-changing technologies in the defense industry. The opinion-forming Gartner includes them in its list of ten strategic technologies that can streamline corporate decision-making processes. When Porsche and Volkswagen Group reach for them, it’s a signal for the automotive industry to  become interested in digital twin technology for good.

Although the concept of a digital twin had been developing in the space industry since the 1970s, it was not until the 1990s that it was first mentioned in the literature (the book entitled  Mirror World, by David Gelernter). In industry, the technology was recognized even later, 30 years after the Apollo mission, when the authority in the field of PLM - Michael Grieves - disseminated it.

Today, the technology, which was officially named Digital Twin by NASA just over 10 years ago, is placed on the pinnacle of key solutions at the convergence of the virtual and real world. It works well wherever there is a high number of failures, work of coupled systems, and where the production process is long and burdened with numerous risks.

The  automotive industry is one of these sectors, as demonstrated by the virtual car concept developed by Porsche for the new Taycan. What is a digital twin and what benefits does it bring?

•  test prototypes for their functionality, durability and user expectations;
•  predict defects and analyze possible design errors;
•  save time and financial means;
•  reduce design and production risks;
•  improve monitoring capabilities of vehicle fleets;
•  and best of all - it enables continuous product improvement, as it often collects data from not only one, but thousands of objects. This makes it learn faster and predict defects more precisely, as it is based on knowledge gathered from a vast  number of sources.

In the case of cars, these could be sensors from dozens of systems spanning the entire vehicle lifecycle: from research and development to the manufacturing plant and  OTA updates , to  connected services .

„Chassis twin” - a virtual car concept developed by Porsche for TaycanThe „chassis twin” project has been in the process of development at Porsche for the past three years and then it was continued by the CARIAD company (the Volkswagen Group's vehicle software provider). The air suspension of the new Porsche Taycan was chosen as the main object.

Why the chassis and not the entire car? Because it is this part of the vehicle that is subjected to the most strain, especially on racetracks.

Porsche engineers used intelligent neural algorithms to centrally analyze the data, and in-car sensor data was collected not only from Porsche cars but also from Volkswagen Group vehicles. This increased the data pool by over 20 times. The "chassis twin" concept enables chassis loads to be detected, even if they are not noticeable inside the cabin, and notify the driver before faults appear, even when no suspicious sound or vibration has yet been noted by the driver or mechanic.

The data collected by the vehicles is sent via Porsche Connect to a  central system in the cloud , where an algorithm calculates the relevant durability and vehicle operation thresholds for the whole fleet of vehicles to create a baseline. When these are possibly exceeded, the driver of a specific vehicle receives a notification via Porsche Communication Management (PCM), that the chassis may require inspection. Looped into the computational work, the algorithm recommends not only the type of service needed but also the scope of work to be carried out in the service center.
By removing a faulty or overloaded component early on, the driver will not only be able to avert potential malfunctions, but also keep the vehicle in better overall condition.

In the future, based on a vehicle's digital usage history, Porsche or a  partner insurance broker can offer extended warranties and better services to the driver . The data can be classified, analyzed, and used not only for repairs to a specific vehicle but to predict life-cycle events for the entire product. This allows Porsche to create new services and features, and to test different scenarios for the development of a particular vehicle line. Thus, it saves the time and resources required to bring ill-conceived solutions into reality.

The other possible use case is that drivers themselves can use the  data collected by the digital twin to negotiate with the prospective buyer. The buyer can view the vehicle's overall condition and chassis service history.
As for drivers' concerns about their own privacy, the manufacturer assures that it collects data anonymously and the system does not store any information that could identify the driver directly. The future of the concepts digitization of the automotive industry is gaining pace year after year. The digital twin concept may prove to be one of the key technologies that will push  software-defined vehicles to new tracks and help companies create safer cars, provide new services and increase vehicle lifespan.

So far, half of the Taycan users have signed up for the Porsche pilot program, which collects data from the chassis of their sporty electric cars. In 2022, the program is to launch at a test level, and only sensor data directly from the mechatronic components will be evaluated. In the future, the concept is to reach its full potential, making it possible, among other things, to calculate the wear and tear of specific components without the need for physical measuring devices.

How will the  "chassis twin" model developed by Porsche work out ? The future will tell. What is certain is that a return to the past is only possible in the movies. In 2022, Volkswagen is commencing an era in which a virtual equivalent will soon await the driver, in addition to their actual real vehicle.

 While the automotive industry is rapidly changing by adopting a software-first strategy, like in other sectors, automotive enterprises struggle with productionizing AI and ML R&D projects. Machine Learning and Data Science teams face numerous challenges, including determining the proper technology, automating workflows, managing computing resources, managing data, and building solutions meeting internal regulations. All these issues can complicate the project even before the kick-off.

So, how do we support AI teams to overcome typical challenges and enable ML engineers and Data Scientists to focus on creating and bringing artificial intelligence algorithms to production?

The implementation of a dedicated deployment platform is a solution that is well suited for  the automotive industry . In particular, it allows you to:

•  accelerate the productionization of AI and ML applications;
•  provide an easy and quick project and user onboarding;
•  simplify access to data and computing resources;
•  ensure high scalability -even when the number of accounts far exceeds thousands of users.

To illustrate the process of working on the platform, let's have a look at a project that the Grape Up expert team had the opportunity to implement.

Building AI and ML deployment platform using proven cloud-native technologies - practical use case

Our client - a well-recognized sports car manufacturer - set us the goal of designing a reliable and extensible architecture capable of handling hundreds of customer accounts for the platform. Tools were to be selected for the project to ensure the scalability and flexibility of operations. The idea was to provide fast and efficient  production of AI/ML software .

Along with building the platform architecture leveraging Terraform orchestrating Cloud Formation scripts, Grape Up ensured efficient migration of existing environments. The solution was integrated with Continuous Integration pipelines and the E2E tests set. To reap the benefits of high-quality performance in multiple regions worldwide, the platform was hosted on the AWS cloud.

Results?

An  AI Deployment Platform was delivered , which was capable of managing a huge number of AI/ML projects and allowed for streamlined processes to create, test, and deploy artificial intelligence and machine learning models into production for  Data Science teams.

Developers were guided through the company's deployment processes and supported with reusable blueprints that could be leveraged at the initial steps of the development.

The cloud-native toolkit that was created provided flexibility and agility, at the same time supporting innovation in the vendor's operations. After introducing improvements to the platform, the customer could reduce the code by 80%, while retaining high quality and testability.

All those solutions allowed  AI software development teams to work more efficiently and reduce time-to-market for new products and services.

 The subscription business model outperforms the non-subscription business as we observe the shift from ownership towards usership. The success of services like Netflix for movies, Spotify for music, Microsoft Game Pass for video games quickly spreads outside from the entertainment industry. For industries based on the services, not on physical products, the shift from one-time or reoccurring payment to subscription-based service is easier.

Not-so-warm welcome for subscription model in automotive

It’s harder to justify monthly payments for the service if it enables the device or feature already present in the product you have already paid for. The huge backlash in media happened in a reaction to BMW’s announcement in 2020 that the company has been considering selling heated seats as a subscription-based feature.

Most of the criticism was associated with the fact that the overall price of the car was the same, and obviously, it was already equipped with the electric pad in the seat, while there was no additional software for handling the feature - it was just enabling and disabling the button. Some of the journalists even called that “simple-feature-as-a-service.” That resulted in nightmarish marketing for the idea and postponed the implementation of similar models by other OEMs,  encouraging them to rethink the risk of making similar announcements.

Subscription business models become new normal

But in the last few years, a lot has changed. People get used to the subscription payment model appreciating its benefits. Also, more and  more features in the car are based on the software , so it’s just easier to justify additional payment over the initial vehicle purchase. The overall reception of the idea shifted from mostly negative to neutral or positive.

At this point, customers understand that sometimes building-in hardware that is not activated can be cheaper than making hundreds of configuration versions. Also, the vehicle manufacturers start to learn how to better advertise the benefits of the subscription business model and how to better pick the features that fit this kind of model. It also enabled certain flexibility if this is a corporate vehicle, with basic default configuration, and the driver would like to add Apple CarPlay or enable additional convenience features, like opening the car with the smartphone.

How OEMs can successfully implement a subscription model

For OEMs planning to start with this business model, the moment for building a platform supporting subscription-based services is now. Not just the payment system, because the subscription business model is based on the idea that the feature can be enabled and disabled on the fly, basically  making Connected Car a key requirement , while  OTA makes it much more robust long-term .

It all starts with requirements. Let’s try first to distinguish typical types of feature purchase for a vehicle:

•  Standard equipment (automatically activated in production phase)
•  Runtime - associated with the driver or limited-time licenses
•  Lifetime - associated with the vehicle, does not expire  
   
  •    Purchased on an initial configuration in a dealership  
   
  •    Purchased aftersales (either in dealership or online)  
   
•  Subscription - associated with driver or vehicle, can expire  
   
  •    Automatic re-subscription (e.g., no end time)  
   
  •    Manual re-subscription (e.g., end time after X months)  
   

The other key aspect is offering differentiation between countries, regions, and continents. The same feature may be available as a subscription in the EU while only available as a one-time purchase in the USA.

To make the offer complete, the manufacturer may allow buying a custom offer specific to geographical location - for example, the additional package offered when the driver enters Nürburgring - 24 hours of additional racing time-tracking features.

Building a system for the new model

Our system has to handle all those use cases. To accomplish that, we need to build a solution in which the scheduler (sometimes called cron, from the name of Linux job scheduler) is the core component. It is responsible for triggering notifications or events at specific periods - for example, resubscription notification monthly, to trigger the payment, or license cancellation event after a configured period.

The scheduler itself is just a single, small part of the system. The other important piece is the database for storing the subscription status and the API backend for retrieving and updating the values. Consistency is crucial, as the feature getting disabled by mistake leads to a bad user experience.

 The system has to be connected to the vehicle . In most cases, this is done asynchronously through queues like Kafka or RabbitMQ. This gives better stability and reliability than direct connections.

Lastly, we need to ensure that the feature is actually enabled or disabled in the vehicle. This means the vehicle has to receive the correct, unique license for that feature when it’s enabled, and revoke it when it is disabled (alternatively, the license can be automatically pushed every, for example, 30 days, with expiration time set to 33-35 days, to prevent feature loss when connectivity or payment problems occur.

To avoid building an additional retry mechanism into the licensing system itself, it’s better to update the feature state using Digital Twin. In this case,  the digital representation of the vehicle is updated with the new license, and it is then responsible to synchronize itself with the vehicle when the internet connection to the car is available. This makes the system conformant to the single-responsibility principle, so the license system does not have to know or understand the vehicle connectivity.

That’s the basic architecture of our system for handling licenses and subscriptions of digital services. Obviously, that’s just the beginning. For OEMs, where the scale of digital business grows exponentially, the next important topic would be the reliability of the system. For that, scaling to meet the demand is important, as well as caching the current state of licenses to avoid complex queries.

Apart from that, this is enough to start with the feature activation and deactivation and handling subscriptions. Of course, it must be connected to mobile apps and online stores for purchases and to payment systems, but those are already used by most enterprises.

Is this really the future? It seems like we can’t avoid it anymore, especially with  shared mobility growth, the ability to unlock temporarily additional features is tempting. Imagine grabbing a Ferrari for a weekend to take it to the track, enabling an additional 50HP and an  advanced AI for measuring your times and proposing a better moment to break before the turn and accelerate afterward. And paying for only 2 days. This may make all the difference in convincing customers to the new subscription business models.