Dataset Construction for VTLA Foundation Model through Tactile Data Collection in Manufacturing Environments Selected for METI/NEDO Public Project

Three robot manufacturers and universities/sensing companies collaborate to accelerate the social implementation of physical AI.

Kawasaki Heavy Industries, Ltd., Osaka University, FANUC CORPORATION, FingerVision Inc., and Yaskawa Electric Corporation have proposed and been selected for the "Post-5G Information Communication System Foundation Enhancement Research and Development Project / Research and Development on Data Ecosystem Construction, etc. (GENIAC)" (hereinafter referred to as "this project"), a subsidized project solicited by the Ministry of Economy, Trade and Industry (METI) and the New Energy and Industrial Technology Development Organization (NEDO).

This project aims to promote the implementation of physical AI in manufacturing environments. Three leading Japanese robot manufacturers, along with universities and companies in the sensing and IT fields, will collaborate to develop a "VTLA (Vision-Tactile-Language-Action) model" that can comprehensively handle data related to vision, tactile, language, and action collected in manufacturing settings.

Furthermore, by designing, collecting, and accumulating datasets suitable for the VTLA model and constructing a data ecosystem, we will achieve the automation of complex and delicate manual tasks that have been difficult until now.

The implementation period is scheduled for one year from August 2026 to July 2027. We will also collaborate with companies and research institutions such as ABEJA Inc., the National Institute of Advanced Industrial Science and Technology (AIST), and Nagoya University, National University Corporation Tokai National Higher Education and Research System, to advance foundational technologies.

Currently, the Japanese manufacturing industry faces challenges such as a decrease in skilled workers and the need to respond to increased sophistication and diversification of production. Digitalization of advanced tasks that are difficult to handle with conventional automation technologies is required. In particular, automation through the application of AI to tasks involving non-visual information such as tactile and force sensing is expected.

Meanwhile, as international competition in the AI and robotics fields accelerates, leveraging the high-reliability, high-quality manufacturing data that Japanese manufacturing has accumulated over many years is considered key to strengthening industrial competitiveness.

This project will address these challenges by developing a data foundation and model technology that integrates multiple sensory information, including vision and touch, thereby realizing the reproduction and automation of advanced tasks in manufacturing environments and contributing to the strengthening of industrial competitiveness.

Key Points of This Project

Three robot manufacturers will jointly participate to standardize data specifications and collection infrastructure, building a dataset usable with various robots and devices.

Development in a short period, considering the speed of technological evolution, and early formation of a data ecosystem.

Demonstration of the VTLA model through collaboration with startups and universities possessing expertise in the tactile information domain.

By promoting this project as a unified effort by the robot industry, we will accelerate the adoption of robots not only in manufacturing but also in various other fields, contributing to the solution of social issues such as the declining working population. Furthermore, even after the completion of this project, we will promote the expansion of the data ecosystem based on the results achieved, contributing to the advancement of the entire robot industry.

<Reference>

METI Press Release Regarding This Matter

https://www.meti.go.jp/press/2026/07/20260702001/20260702001.html

Inquiries Regarding This Matter

Kawasaki Heavy Industries, Ltd. https://www.khi.co.jp/corporate/contacts/

Osaka University (Graduate School of Engineering Science, Harada Laboratory) TEL: 06-6850-6380

FingerVision Inc. MAIL: [email protected] TEL: 080-5348-2167

* Information contained in this press release is current as of the date of announcement.

Please note that it may be subject to change.

◆Company Overview

- Company Name: FingerVision Inc.

- Representative Director: Yuki Konno

- Locations:

- 1F, Mitsui Woody Building, 2-4-14 Toyo, Koto-ku, Tokyo

- 1F, Light One Building, 73-1 Suzaku Hozou-cho, Shimogyo-ku, Kyoto City, Kyoto Prefecture

- 2F, NK Marunouchi Building, 2-17-13 Marunouchi, Naka-ku, Nagoya City, Aichi Prefecture

- Establishment: October 2021

- URL: https://www.fingervision.jp/

- YouTube: https://www.youtube.com/channel/UCRgZuTJvDxjD_7ta03BRQFA

- LinkedIn: https://www.linkedin.com/company/fingervision/

- X/Twitter: https://twitter.com/Yuki_Nono_fv

- Email: [email protected]

- Webinar (FingerVision Robot Explanation): https://tv.aperza.com/watch/2069

◆Technology Overview

Our core technology is based on the concept of "reproducing tactile sensation from images (cameras)." By mounting this on the fingertips of robot hands, etc., robots can perceive tactile information (force distribution, slip distribution, etc.), enabling control that allows robots to handle objects as if humans were using the sense of their "palms." It is highly practical, featuring high functionality (high resolution, multi-modality) while being economical. Although called a "tactile" sensor, it is a completely new concept of "visual-tactile sensor" that also incorporates visual modality to see the grasped object, possessing infinite application possibilities beyond process automation combined with robots.