Digital disruption is changing the automotive industry in a big way, be it connected vehicles or autonomous vehicles.

Developing and testing autonomous vehicles (AV) requires a highly sophisticated collaborative platform that can enable various autonomous building blocks to work together. Different users like data labelling teams, data scientists, testers, modellers, and hardware teams, who come together to accomplish the objectives, need a robust platform to manage rich and complex data pipes.

Challenges in developing and testing autonomous vehicles

Auto companies face numerous challenges for their AV Dev / Test requirements. The challenges include:

• Handling ever-growing data storage needs
• Optimizing investment in on-premise data centres
• Hybrid cloud support
• Regional regulatory restrictions
• Catching up with advancements in AI / ML ecosystem
• Handling petabytes of test data spread across different test locations

Beating the challenges with a sophisticated collaborative platform

Autonomous Vehicles Remote Collaboration Platform (AV RCP) with its distributed architecture combined with Microsoft Azure services addresses most of the testing and developing related challenges of the AV industry described above.

Figure 1 shows the high-level architecture of AV RCP. The distributed architecture helps in processing the data at the source instead of pushing all the data to a centralized location. This helps to ensure data does not cross geographical boundaries for regulatory compliance and minimizes huge data transfer costs. The metadata extractor helps in quickly searching and navigating through the large data files and extracting the relevant information that is of interest to the user. The on premise deployment option helps in optimizing the existing cost and the hybrid cloud option provides support for leveraging cloud capabilities and expansion plans in the future.

Autonomous test fleet cars collect ground truth data through various sensors while driving through city roads. Once a test vehicle reaches test centres, on-board data stores upload data to the server hosted locally at test sites. Test engineer cleans (semi-automated) data files and uploads it to the local shared file location, to enable access to all users across the globe.

Once the file is available in the shared file location, AV RCP’s Optima metadata extractor extracts meta data of the file. It also provides a facility to assign user tags. Metadata information is stored in the central database, which is searchable by any global user. Let’s say, an user from the India lab searches for some data file and finds that it is available in the US lab.

There are several ways the user can access this file. User can download the file to local for processing or have a mechanism where they can perform required operation without downloading file locally. Downloading the file is straightforward but requires huge network resources. Here, file size is in gigabytes to petabytes. Downloading such huge files will consume precious network resources and time. Typically, a 50-60 GB file takes around 5-6 hours over a regular network.

AV RCP distributed architecture provides a different approach without the physical movement of the file. User clicks on the desired file; system redirects VDI enabled desktop (for Azure Cloud, Azure Desktop is an option) where user can access and process file remotely. As the data file is available locally to compute engine (remote), simulation and other data processing can be done seamlessly.

AV RCP solution also provides tools to slice files so that user can download the data-of-interest, typically a smaller size file locally for any hardware simulation/ test. This approach saves precious network bandwidth and saves on file upload and download time. It thus improves productivity of the data engineer.

AV RCP can be deployed in hybrid and cloud PaaS environments:

• Hybrid: AV RCP supports Hybrid cloud deployment model. In this model, high computation (Simulation, AI / ML) and storage archival are leveraged from cloud based services and the rest of the functions are retained at on-premise lab infrastructure. This is the recommended deployment. Many Azure cloud services can be leveraged for scalability and performance in this option.
• Cloud PaaS: AV RCP is available as a PaaS service. It exposes RESTful APIs so the customer can use RCP platform services in any application. Due to multitenancy feature, AV RCP can support multiple customers.

AV RCP framework uses various Azure PaaS / IaaS services for achieving higher scalability and reliability (See Table 1).

Many countries have data movement restrictions where data cannot move out of the jurisdiction of a specific geography. AV RCP provides a unique solution for this restriction using its distributed architecture. Global developers can work with data without moving it from specific regions.

The business benefits of AV RCP include:

• Retaining existing investment on private data centre and provision to leverage cloud for future business expansion plans
• Better utilization of available network bandwidth
• Quick turnaround time for accessing remote resources
• Enabling Dev, Test and Data Scientists to use resources across global test labs
• Higher level of scalability, reliability and security due to its seamless integration with Microsoft Azure ecosystem including Azure AI / ML and ADAS services
• Adherence to regional regulatory compliance on data movement

Racing ahead with AV RCP

The distributed architecture adopted by AV RCP is a progressive approach and provides a cost-optimized environment that is highly responsive to the needs of the users. The solution utilizes existing on-premise data centres and uses Microsoft Azure for extensibility and scalability. It grows seamlessly as the user and storage volumes grow and adapts to the future business needs as well. The real value of the architecture, design, and application lies in the manner it delivers superior user experience.

Akhil Gokhale

Managing Consultant – Scalable Platforms and Software Products, Industrial & Engineering Services, Wipro Limited.

Akhil has around 18 years of experience in product and software development. He currently leads development and architecture of cloud solutions and accelerators using open source products and leads large digital transformation programs.

Radhakrishna Singuru

DMTS - Senior Member – Scalable Platforms and Software Products, Industrial & Engineering Services, Wipro Limited.

Radha has around 24 years of experience in product and system software development in varied areas, ranging from cloud and virtualization technologies, scalable platforms, SDN, L2/L3 switching and stacking software, etc., across multiple industry domains. He currently heads the Scalable Platforms, Software Products and Payments Practice.