Introduction

In traditional manufacturing industries, research and development (R&D) capabilities have long been one of the key factors in gaining a competitive edge. However, recent advancements in digital technology have ushered in a new era. Today, corporate competitiveness depends not only on strong R&D capabilities but also on an enhanced R&D environment that maximizes these capabilities. Even with equivalent R&D capabilities, an environment that uses digital technology to support rapid trials and errors (i.e., high cyclization) can significantly improve both the quality and speed of outcomes. Nowadays, building cloud-native and AI-native R&D platforms, along with creating environments that enable researchers to generate outcomes at high cycles, is a particularly important key to corporate competitiveness. This paper illustrates this transformational shift by referring to two specific examples of R&D environment development at OMRON: one focused on cloud development and the other on the utilization of generative AI.

1. RDinX Cloud Development Environment

1.1 State of the cloud development environment for achieving high-cycle R&D

OMRON operates across various business areas, ranging from factory automation (FA), which automates production processes, to healthcare, devices and modules, social systems, and energy solutions. OMRON's R&D headquarter is responsible for high-technology development within these areas. In the power electronics area, particularly for FA and energy solutions, we have actively adopted computer-aided engineering (CAE) and optimization technologies for thermal design optimization. This shift has moved our design process away from traditional reliance on experienced designers and repetitive prototyping, resulting in a significant reduction in the design workload.

Nevertheless, despite the fact that using CAE and optimization technologies requires a substantial amount of high-performance computing resources, traditional hardware-dependent R&D environments, including personal computers (PCs), remained the mainstream on the R&D shop floor. Under these circumstances, it could take several months of lead time to procure large quantities of physical PCs and get ready for experiments with each PC requiring a manual environment setup. Consequently, challenges emerged regarding the considerable time and workload required to initiate research and development.

1.2 Cloud development environment building efforts at OMRON and the challenges

To solve the aforementioned challenges, OMRON's R&D headquarter began virtualizing high-spec PCs using the cloud platform provided by Amazon Web Services (AWS). Cloud-based virtual environments enable researchers to obtain the required amount of computing resources in almost real time as needed, allowing for timely experimentation as desired.

However, virtualizing R&D environments required a change in mindset, which involved transitioning from on-premises R&D environments hosted on the local network to cloud-based virtual spaces. Thus, we faced two major challenges in achieving this transition.

The first challenge was security design. The experimental data and analysis results generated from R&D processes are valuable technology assets that form the source of OMRON's competitiveness and are highly confidential. In conventional on-premises environments on the local network, a robust security design developed by the headquarters IT department protected these technology assets. Transferring them to the cloud required maintaining an equivalent or higher level of security.

The second challenge was cost optimization. Cloud-based virtual PCs are billed proportionally to the specs and usage time on a pay-as-you-go basis. Without proper planning, the usage cost can rise quickly. Therefore, it was imperative to develop a scheme to manage usage properly and keep the cost down without reducing the latitude and flexibility of research and development.

OMRON's R&D headquarter addressed these challenges by implementing the following measures to establish the R&D Infrastructure on Cloud Systems (RDinX), an environment designed to balance convenience for researchers with improved quality and speed of research and development.

1.3 Measures taken to address the challenges

This subsection describes the two aspects into which the advantages of RDinX can be summarized.

1.3.1 Measures taken for safe security design

The first aspect is a security design strictly compliant with OMRON's information security rules. Before starting the development of RDinX, we conducted a detailed analysis of the matters regulated by our information security rules and integrated them into the conceptual design. Adopting the balance between convenience and security as the design concept, we aimed to deliver RDinX as OMRON's R&D DX platform. While urged to meet the demand for early delivery of a user-friendly cloud environment that drives research and development, we were concerned that the security level might be compromised when multiple teams accessed the environment and that delays in service speed might occur because of the increased operational workload. AWS Inc. officially recommends preparing more than one management unit, referred to as an "account," for each intended use when building an AWS environment.¹⁾ Here, we faced the technical dilemma of ensuring standardized security settings without compromising the speed of scaling up RDinX usage.

We adopted the strategic utilization of the AWS Control Tower service²⁾ to solve this challenge. This service enables the efficient management of multiple accounts, leveraging best practices established by AWS in its experiences of collaborations with several thousand enterprises. A particularly decisive factor for us in adopting Control Tower was that it enabled rapid standardization of security settings through the active use of centralized distribution of security settings from a representative account to all others. Prior to the implementation, we scrutinized the GuardRails³⁾ offered by AWS and adopted the most suitable ones as cloud-specific security rules. We performed a detailed analysis of our company-specific security requirements and systematized them as cloud environment settings.

This scheme worked as expected, enabling us to deliver accounts complete with security settings to the R&D headquarter or collaborative creation partners within short lead times. This design, which embodies the concept of the balance between convenience and security, is highly valued for its innovative nature as can be seen from the fact that AWS Inc.'s official website featured it as the first case of the full-scale use of Control Tower in Japan.⁴⁾ We built a dedicated portal site for researchers in the R&D departments. When a researcher needs a virtual PC, they can start using it in only approximately five minutes just by accessing this portal site and specifying the parameters, such as the required CPU/GPU power and memory capacity. The security settings mentioned above are automatically applied to the virtual PC to be delivered, allowing the researcher to focus on their research activity in a secure environment without needing to be aware of the security settings. Additionally, virtual machine image cloning, including environment settings, eliminated the duplication of setup work involved in introducing multiple virtual PCs.

1.3.2 Measures taken for cost optimization

The second aspect is a system design and an operational process aimed at cost optimization. The most important factor in achieving cost optimization is an accurate understanding of actual usage. In other words, it is crucial to determine when researchers should use how many virtual PCs of what specs at what frequency for how long.

To clarify these variables, OMRON's R&D headquarter put RDinX into trial operation and analyzed the actual usage history of virtual PCs in detail. The analysis results showed that purposes other than the execution of AI or optimization training only require a limited rather than a huge amount of virtual PC resources. Additionally, we also confirmed that because cloud-based virtual PCs are charged proportionally to uptime, assured power-off during non-use leads to effective cost reductions.

Typical cloud usage cost reduction measures include utilizing volume discounts through the consolidation of in-house accounts and cost reduction options available from AWS Inc., such as the Reserved Instances⁵⁾ and Savings Plans.⁶⁾ However, the usage status analysis results for research and development at OMRON have revealed that assured power-off during non-use more significantly contributes to cost reductions than such measures.

Based on these findings, we developed our proprietary virtual PC management console. This console has enabled users to fine-tune settings, such as running the virtual PC only for the specified hours or auto-powering off at specific hours. As a result, researchers can now easily set flexible operation plans to suit the needs of the research process, such as auto-power off on non-working days and at night, or to basically turn the power off on non-working days and at night but leave it on the forthcoming Saturday and Sunday to run the learning algorithms, for example.

Moreover, we also implemented additional functions that enable effective cost management while minimizing the burdens on researchers. Examples of these functions include the following: a function to visualize virtual PC usage time and fees per researcher, and a reminder function that automatically sends an alert email when budget consumption exceeds a certain ratio.

1.4 Effects of introducing RDinX

Through the efforts described above, OMRON's R&D headquarter built a DX platform named RDinX, thereby solving the two essential challenges of introducing cloud technology: security design and cost optimization. This action significantly reduced the workload required for procurement and environment setup during the R&D processes. As a result, researchers can now focus on creative activities per se, dramatically improving our R&D competitive edge, which is the exact form of achieving our intended R&D digital transformation.

2. Generative AI utilization environment of RD Buddy

2.1 Current state of generative AI utilization and the challenges

In November 2022, as we promoted digital transformation through the building and operation of RDinX, OpenAI, Inc.'s release of ChatGPT marked the beginning of a new technological trend of generative AI.

Following the advent of ChatGPT, we received numerous requests from researchers in OMRON's R&D departments, urging us to develop, as soon as possible, an environment that enables the in-house use of generative AI, which effectively improves the efficiency of R&D processes. In response, our department started a detailed investigation toward the introduction of such an environment. However, as with the case of RDinX, two essential challenges emerged: security design and cost optimization.

The primary concern in security design was the risk of information leakage. Researchers may unintentionally enter confidential information into prompts while using generative AI services. The use of such information by service providers for retraining for model enhancement may lead to the leakage of corporate secrets. Suppose, in particular, that personally identifiable information is contained. In that case, such leakage may constitute a case of extra-purpose use or transfer to third parties regulated under the Personal Information Protection Law, resulting in the risk of legal compliance violations.

The cost optimization-related challenge stemmed from the billing system of generative AI services. Typical generative AI services adopt a pay-per-use billing system, which charges proportionally to the number of prompts input or output. Hence, for purposes that involve handling large volumes of technical information, such as research and development, increases in the use of these services tend to increase the cost. Especially when researchers move beyond the phase of manually operating the chat interface and start calling up generative AI from a system automated via an application programming interface (API), a huge amount of input and output information will highly likely occur quickly, causing the cost to skyrocket.

2.2 Building the generative AI utilization environment

To address the challenges described above, we undertook building our proprietary generative AI platform, RD Buddy, specifically designed for R&D processes.

The advantages of RD Buddy can be summarized into the following two aspects:

The first aspect is a multilayered security design for preventing information leakage. RD Buddy is hosted on an OMRON-dedicated private cloud environment built on the RDinX platform, in other words, the AWS, thereby minimizing the risk of confidential information leakage to external services. The first layer is built with its dedicated resources being cloud services (Amazon Bedrock and Microsoft Azure OpenAI Service) that provide generative AI models. This layer ensures complete separation from public generative AI services shared with general users.

The second layer features a content filter, which blocks attacks that use requests to generative AI models, especially prompt injection attacks.

In the preceding stage of the third layer, an API processing layer is provided to handle requests to generative AI. This layer provides external attack blocking, authentication and authorization, and log monitoring, thereby enabling secure operation.

In the fourth layer, a function is implemented that allows the user to select a private mode that excludes highly confidential chats from the knowledge database.

This implementation has enabled full compliance with OMRON's strict information security rules. Furthermore, a comprehensive log monitoring function has been implemented to track information entered as prompts, providing a technically secure environment that prevents information leakage.

The second aspect is an efficient server design aimed at cost optimization. The operation of a generative AI model requires a high-spec environment. Hence, we adopted a design that fully exploits the advantages of a serverless architecture, which utilizes only the necessary resources as needed.

We built a mechanism that specifically addresses the lengthy response times of generative AI. In this mechanism, a minimum connection state is maintained while generative AI creates a reply to a user query; actual computational processing occurs only as necessary. The mechanism ensures responsiveness without compromising the user experience while preventing the occurrence of fixed costs due to continuous connection over many hours.

Additionally, we delivered an interface that enables optimal model selection tailored to the specific use case, allowing users to choose models that prioritize either accuracy, response rate or cost-effectiveness—depending on the requirements. This method prevents cost expansion due to the use of unnecessarily expensive models.

Implementing a server configuration thus optimized, we successfully achieved a significant reduction in usage costs across the entire company, albeit that items billed per usage remain. Building RD Buddy, we simultaneously solved two major challenges to the utilization of generative AI in OMRON's R&D headquarter: information leakage risks and the cost expansion problem. Thus, we provided a safe and efficient environment for researchers to utilize generative AI. Currently, OMRON's R&D headquarter is working on specific applied developments based on this infrastructure. These development efforts include developing natural language-based robot control algorithms and our own generative AI-powered intellectual property agent⁷⁾ which compiles Japanese Patent Office information into the database. These developments are respectively under verification for putting into practical use.

3. DX promotion framework and prospects

It is worth noting that RDinX and RD Buddy were built and are operated by insourcing to OMRON SOFTWARE Co., Ltd. (OSK), an IT specialist company within the OMRON Group. This insourcing approach provides more than one strategic advantage.

First of all, OSK has engineers well-versed in OMRON's information security rules and strategically develops human resources with AWS implementation specialist skills.⁸⁾ Hence, they can implement advanced security designs speedily with high quality. The second advantage is that OSK is an intra-group company. As such, OSK has a well-established system for properly handling OMRON's confidential information.

Moreover, the functions provided by RDinX and RD Buddy are converted into assets by function, such as a security platform, a virtual PC delivery platform, and a generative AI utilization platform. This modular design enables the selective introduction of functions required by divisional product development teams, thereby allowing for flexible and efficient system deployment.

Through these efforts, OMRON promotes activities aimed at digital transformation, not only of R&D processes but also of the entire product development process. Notably, the in-house developed generative AI utilization platform technology attracted significant attention from the outside and, as a result, was spun off and released by OSK as a commercial service called koto-Buddy for use outside the OMRON Group in December 2024. This case saliently exemplifies that OMRON's R&D DX efforts go beyond intra-group technological innovations and contribute to creating new business opportunities.

Cases of achieving enhanced development process efficiency and qualitatively improved outcomes through the introduction of the cloud-native and AI-native R&D platforms RDinX and RD Buddy have followed one after another in OMRON's R&D departments and some of its product development departments. OMRON's efforts, as presented herein, foreshadow a major transformational trend that is approaching the entire global and Japanese manufacturing industries. This wave will evolve into a fundamental paradigm shift, transcending a mere change and transforming what research and development ought to be.

However, a true digital transformation of research and development can never be achieved just by passively introducing systems shaped far on the other side of the sea. It never comes to fruition until researchers go through the entire process of identifying the essence of their creative activities and actively creating and utilizing high-cycle R&D-enabling DX platforms through collaboration with a DX platform-building department like ours.

We are now standing at a critical turning point in the history of research and development. The creative utilization of cloud and AI power is unlocking once-unthinkable R&D possibilities. At the opening of this new era, we conclude this paper with the following words:

"We are the ones who change the world!!"

References:

1）

AWS Inc. “Benefits of using multiple AWS accounts.” AWS Documentation. https://docs.aws.amazon.com/whitepapers/latest/organizing-your-aws-environment/benefits-of-using-multiple-aws-accounts.html (Accessed: Mar. 27, 2025).

2）

AWS Inc. “AWS Control Tower.” List of AWS Products and Services (in Japanese). https://aws.amazon.com/jp/controltower/ (Accessed: Mar. 27, 2025).

3）

AWS Inc. “How Controls Work” (in Japanese). AWS Documentation. https://docs.aws.amazon.com/ja_jp/controltower/latest/userguide/how-controls-work.html (Accessed: Mar. 27, 2025).

4）

AWS Inc. “OMRON Builds R&D HPC Infrastructure on AWS, Uses Optimal Computing Resources to Lead Innovative Technology Development” (in Japanese). Cloud Solutions. https://aws.amazon.com/jp/solutions/case-studies/omron-case-study/ (Accessed: Mar. 27, 2025).

5）

AWS Inc. “Amazon EC2 Reserved Instances” (in Japanese). Amazon EC2. https://aws.amazon.com/jp/ec2/pricing/reserved-instances/ (Accessed: Mar. 27, 2025).

6）

AWS Inc. “Savings Plans” (in Japanese). Savings Plans. https://aws.amazon.com/jp/savingsplans/ (Accessed: Mar. 27, 2025).

7）

T. Okuda et al., “Innovating Intellectual Property Management Based on IP Landscape _ Systematization and Practice of Ambidextrous Intellectual Property Activities at OMRON_” (in Japanese), IP Journal, no. 32, 2025.

8）

AWS Inc. “Developing AWS Human Resources To Achieve Group's Digital Strategies - Seven Obtained Five AWS Certificates Through Half-Year Training, Continue In-House Workshop to Build 100-Strong Team of AWS Engineers” (in Japanese). AWS Training Case Study. https://aws.amazon.com/jp/training/case-studies/osk-training/ (Accessed: Mar. 27, 2025).