―Utilizing operational data from Lawson stores to accelerate the social implementation of physical AI―
KDDI CORPORATION
Tokyo, September 9, 2026―KDDI announced today that its proposed project, "R&D of a Robotics Foundation Model to Replace On-Site Operations in Retail and Logistics" ("the Project"), has been selected (
Note 1) for the Generative AI Accelerator Challenge (GENIAC) (Note 2), a national project organized by the Ministry of Economy, Trade and Industry and the New Energy and Industrial Technology Development Organization (NEDO).
The Project focuses on the Lawson S KDDI Takanawa Head Office Store, a location in Takanawa, Tokyo exclusively for KDDI Head Office employees (Note 3). At this location, KDDI will collect behavioral and operational data from the perspective of workers (egocentric data), ranging from storage room to sales floor operations, and perform field tests for robot control.
Moreover, with technical insights provided by ELYZA, a company with a track record in developing Japanese large language models (LLMs), and KDDI Research, an organization that drives cutting-edge research, KDDI will develop a robotics foundation model using egocentric data and build an open ecosystem using state-of-the-art robotics from Japan and overseas.
KDDI will hold multiple field tests with robots from Japan and overseas during the Project and will publish the robotics foundation model technology from FY2028 onward, based on established technologies and use cases. Through these efforts, KDDI will advance the introduction of robots in Japan's retail and logistics industries and contribute to the social implementation of physical AI in Japan.
The Project promotes the social implementation of "AI-Powered Productivity" described in KDDI's Mid-term Management Strategy (Note 4). KDDI seeks to become a front-runner in the coming "AI-Native Society," where AI will be embedded across all industries. In addition to telecommunications and AI infrastructure, we will pursue integrated efforts ranging from development to field testing and social implementation of foundation models.
■Background
- A labor shortage totaling around 1 million people is forecast for 2035, with a shortage of 770,000 in wholesale and retail and 230,000 in transportation and mail (logistics). These two industries support the infrastructure of people's lives, making labor shortages a serious issue (Note 5). The use of physical AI is expected to help overcome this challenge by enabling flexible, human-like operations based on an autonomous understanding of on-site conditions. However, challenges remain in implementing physical AI as a viable form of labor in complicated logistics and retail workplaces, where product details and environmental conditions are constantly changing.
- To achieve the social implementation of physical AI, it is crucial to not only collect a large quantity of high-quality real-world data to train robots to perform complex on-site tasks, but also to respond to diverse changes in products and environmental conditions while developing an AI model that can make sophisticated decisions and carry out delicate movements within limited spaces. Securing general-purpose hardware that can replace manual operations across a wide range of workplaces is another key challenge.
■Project Summary
Centered around the Lawson S KDDI Takanawa Head Office Store, the Project involves collecting data and verifying robot operations in everything from storing products and preparing stock in the storage room to arranging and restocking shelf displays on the sales floor. Convenience store operations are extremely difficult to automate with robotics, as they involve a wide range of products as well as constantly changing circumstances. We will establish the foundations for implementing physical AI at numerous retail sites in Japan, with the aim of enabling robots to perform a variety of tasks throughout the store.
The Project is supported by GENIAC, a project organized by the Ministry of Economy, Trade and Industry and NEDO to reinforce generative AI development and promote its social implementation in Japan.
1. Efficiently building a training dataset by combining the real world with simulations
We will efficiently collect a large volume of high-quality data through a fusion of real-world and simulation data, including high-precision egocentric data collected during operations at Lawson stores.
| Real-world data | Store staff will wear lightweight AI glasses that have minimal impact on their regular operations, allowing us to collect egocentric data on the types of products they look at, as well as how they reach for and pick them up. We will also collect video data from the robots' perspective. We will use this input to train robots to perform human-like actions during tasks. |
|---|---|
| Simulation data | Using high-quality real-world data for training, we will configure an AI to simulate a wide selection of movement patterns for multiple scenarios in a digital twin. This process will automatically generate simulation data with many variations that would be difficult to collect in the real world, thereby dramatically expanding our training dataset. |
2. Developing a robotics foundation model that can adapt to complex workplaces
We will use the training dataset with a globally advanced open robotics foundation model to develop a specialized model for retail and logistics that can deftly control multi-fingered robotic hands with human-like movements. Through these efforts, we seek to enable robots to perform advanced tasks that have conventionally been difficult, such as picking up soft items like snack food bags or items with complex shapes, and restocking shelves without disturbing neighboring products.
3. Building an open ecosystem using state-of-the-art robotics from Japan and overseas
We will hold multiple field tests covering a wide range of on-site operations, using our robotics foundation model on a broad selection of robots developed by leading manufacturers from Japan and overseas, without being restricted to specific hardware configurations. By sharing insights gained from field tests with robot manufacturers, we can support the development of robots that can replace a wide range of manual tasks, thereby contributing to Japan's physical AI industry as a whole.
■Collaborations to Drive the Project Forward
The Project involves cooperation between KDDI Group companies and multiple partner companies from fields such as robot development and AI development to build an open ecosystem. The main roles of each company are outlined below.
| KDDI | Conducting overall management and promotion of the Project. Providing telecommunications, compute (GPUs), and security infrastructure; developing a robotics foundation model; developing technologies to collect and utilize egocentric data; collecting on-site data; designing and executing field tests; and pursuing social implementation | |
|---|---|---|
| KDDI Group | ELYZA | Providing technologies for developing a robotics foundation model based on insights from developing LLMs from scratch |
| KDDI Research | Using image recognition and situational awareness technologies cultivated through years of research to provide technologies for utilizing egocentric data, as well as data augmentation technologies for purposes such as robot trajectory generation | |
| Lawson | Providing the field test environment (Lawson S KDDI Takanawa Head Office Store) and insights into store operations | |
| Supporting companies | Genki Robotics | Providing humanoid robots |
| RLWRLD Inc. | Providing a robotics foundation model (base model) | |
Message from Taiji Yoshida, Senior Vice President and Company President of Incubation Company at Lawson, Inc.
We are delighted to collaborate with KDDI in this initiative. Lawson is promoting the utilization of technology to solve a broad range of challenges faced by stores. We hope that the extensive use of the insights and knowledge gained through this initiative will advance the social implementation of physical AI and help tackle labor shortages and other issues in the retail and logistics industries.
- (Note 1)
- (Note 2) Generative AI Accelerator Challenge (GENIAC)Open in a new window: A project collaboratively organized by the Ministry of Economy, Trade and Industry and the New Energy and Industrial Technology Development Organization (NEDO) to reinforce generative AI development and promote social implementation in Japan.
- (Note 3)
- (Note 4)
- (Note 5)
- *The information contained in the articles is current at the time of publication.
Products, service fees, service content and specifications, contact information, and other details are subject to change without notice.