日本語 
Nexdata has invested a total of 2.5 billion yen to expand its data collection facility to 8,000 square meters, and is also providing ego-centric data collection and datasets for foundational models. This solves the cost and lead time challenges in acquiring real-world data for physical AI.
The evolution of AI technology is shifting from the era of large-scale language models (LLMs) specialized in generating information in digital space to the stage of “physical AI” that directly interacts with the physical world and operates autonomously. In Japan, the market is expanding rapidly due to a combination of labor shortages caused by a declining birthrate and aging population, and the need for automation in the manufacturing and service industries. While conventional generative AI primarily aimed at processing text and 2D images, physical AI integrates environmental recognition using sensors with the physical movements of robots, and is positioned as a next-generation infrastructure that directly contributes to solving real-world problems.
Large-scale data is essential for physical AI development.
In the development of physical AI, it is becoming common knowledge in the industry that the “scaling law” applies, just as it does in LLM. To improve the versatility of the model and the control accuracy in real-world environments, large-scale, high-quality real-world data, including diverse physical phenomena and behavioral patterns that cannot be fully reproduced in simulations, is essential. However, data collection in real space is the biggest bottleneck in the development process due to the combined costs of environment construction, the difficulty of synchronizing and calibrating multiple sensors, and the burden of annotation.
Also Read: Mujin Introduces Subscription Model for Robotics
To address this challenge, Nexdata has invested over 2.5 billion yen to construct a dedicated data collection facility spanning over 8,000 square meters. From data collection within the factory and the collection and annotation of first-person (ego-centric) real-world data, to off-the-shelf datasets compatible with environmental recognition, decision-making, and motion control, we provide comprehensive data solutions to accelerate physical AI development. The cost advantages of large-scale production and readily usable data assets in development environments support shorter lead times and improved real-world adaptation accuracy for physical AI and VLA model development.
Large-scale, low-cost data supply realized by an 8,000 m² data collection plant.
Nexdata has invested over 2 billion yen in developing a data infrastructure specifically for physical AI development. Currently, it operates two large-scale data collection plants, with a dedicated space of over 8,000 square meters where it can operate more than 400 diverse robot platforms in parallel, including humanoid robots, quadruped robots, industrial robot arms, and multi-fingered manipulators.
The facility features a variety of scenario environments that faithfully reproduce real-world operating environments, including homes, pharmacies, manufacturing lines, and logistics warehouses, and is staffed by over 600 operators and management personnel.
This enables the efficient production of high-quality physical AI data that covers the entire development phase, from pre-training of large-scale infrastructure models to fine-tuning for specific tasks and imitation learning that learns by mimicking human performances.
SOURCE: PRTimes
English