Nissan Motor uses AWS for data-driven Automotive development, greatly improving development efficiency

Since its establishment in 1933, Nissan Motor has developed pioneering automobiles and services. The company's R&D department is engaged in "data-driven development," which shortens development periods and enables innovative technology development by integrating data accumulated at each phase of automobile development.

The company is focusing on "electrification" and "intelligentization" to deliver attractive and sustainable vehicles. One aspect of intelligentization is autonomous driving technology, which enables hands-free operation using sensor data. The autonomous driving technology "ProPILOT 2.0," announced in 2019, enables hands-off driving on highways, lane changes, and natural control linked to navigation by analyzing a larger volume of sensor data.

"As vehicles become more advanced, the number of sensors will continue to increase significantly, control systems will become more complex, and reliance on software will grow," says Daisuke Tawaramichi, General Manager of the Data Science Group at Nissan Motor's R&D Customer Performance & Vehicle Performance Engineering Division.

Softwareization also increases the ratio of coding, data analysis, and software DevOps/MLOps, so data-driven development is difficult in an on-premises environment where data capacity and resources are limited. To address this, the company has built an integrated development PoC environment (Engineering Workbench) on AWS, enabling access to the necessary development data whenever needed. Engineering Workbench provides the necessary development tools through a unified UI, allowing automotive engineers to access and utilize the software and data required for their work.

“The cloud is essential for analyzing the large amounts of data and gaining insights. Through the Engineering Workbench UI, engineers can leverage cloud technologies and open-source tools directly linked to vehicle development tasks. It is also effective for sharing deliverables that engineers previously managed on their personal PCs and for strengthening governance through access logs," says Tawaramichi.

Nissan Motor Co., Ltd. has built an R&D experimental platform utilizing data in the automotive development test process on the Engineering Workbench.

“The test process, which is the final stage of development, generates a large amount of data. Furthermore, there are multiple analysis software for performing collision tests and aerodynamic performance tests. Since it is difficult to link them in an on-premise environment, we have built an R&D experimental platform that links with APIs using AWS.” (Mr. Tawaramichi)

The R&D experimental infrastructure combines AWS managed services and consists of five functions of data acquisition, post-processing, accumulation, analysis, and visualization from vehicles with microservices. By making these templates, they can also be applied to other experiments.

“The point of the architecture is to upload the generated data directly to the cloud. Routing through local servers and other intermediaries incurs a lot of resources and management costs, so we have structured the functions for directly importing data and visualizing it using microservices." (Mr. Tawaramichi)

Since AWS continues to release services specialized for automobiles, we expect that utilizing these will help reduce development costs and shorten development timelines.

“We can shorten the design process by combining standardized services because the reference architectures specific to automobiles are provided. When I request such a feature from AWS, it is very helpful that they can actively adopt it and make it functional in the next update.” (Mr. Tawaramichi)

By utilizing AWS, it is now possible to handle large amounts of data, and efficient analysis is realized. By automating workloads, human resources have also been reduced to a minimum. As understanding of the cloud has increased within the company, there has been growing interest in exploring new types of analysis, which is one of the benefits of using AWS.

“We've been training our entire R&D department on AWS for 3-4 years. Cloud computing, once considered just a "place to store data" by automotive engineers, has seen increased understanding of AWS services, leading to a broader adoption across the team. In the future, we will standardize the entry point for AWS usage within the R&D department, while accumulating use cases of experiments conducted by engineers and turning them into templates. We will adopt the generative AI for analysis and aim for further development of data-driven development.” (Mr. Tawaramichi)

The R&D department is actively adopting generative AI to improve productivity and create new value, and is continuing various trials and errors on AWS.

"The benefit of using AWS for generative AI lies in the fact that it covers even the UI for business users. There are plenty of managed services for developers to deploy code, making it extremely easy to use.” (Mr. Tawaramichi)

One example of generative AI in R&D is the "AI Data Assistant (ADA)," which enables advanced data analysis in natural language, with multiple APIs built on AWS. As a result, business departments can now perform analysis with chatbots without having to rely on data science teams. Ye Junting, Assistant Manager of the Data Science Group, Customer Performance & Vehicle Performance Technology Development Division, R&D Division, says, “Exploratory data analysis performed by the business department is 3 to 5 times faster than before, and the PoC period has also been shortened by about 80%.”

The second application case is automatic evaluation of search extension generation (RAG) combining large language models (LLM) and search. In the automotive industry, many systems are directly linked to safety, and strict quality control is also required for AI systems. To address this, a system to automatically evaluate the accuracy of the RAG system was built on AWS for the Japanese Q&A Bot of the user manual developed using generative AI.

The evaluation of RAG is carried out in three steps: (1) automatic creation of evaluation Q&A using LLM, (2) review of Q&A by domain experts and creation of a high-quality Q&A list, and (3) automatic evaluation based on the Q&A list along multiple axes. On AWS, the system is built with three components: the RAG system to be evaluated, the backend for automatic evaluation, and the result verification and Q&A review.

“With the RAG automated evaluation system, we can now confirm with certainty whether improvements have been made. Evaluation time and evaluation costs have been reduced by 90% compared to manual evaluation, and verification results are now more than 90% consistent with expert judgment.” (Mr. Ye)

Nissan Motor plans to actively address the challenges associated with the adoption of generative AI and rapidly advance in-house development technologies.