Japan’s GENIAC program has unveiled a new round of projects to develop artificial intelligence for robots, in a step that shifts the program’s focus from general generative AI models toward the data, hardware, and models robots need to interact with the physical world. The announcement was made during an event held in Tokyo on September 9, 2026, with support from Japan’s Ministry of Economy, Trade and Industry and the New Energy and Industrial Technology Development Organization (NEDO).
The program includes a project to develop foundation models for robots, with 13 organizations participating in a working group, as well as a research project to build a data ecosystem involving nine organizations. The organizations participating in the data-ecosystem project had announced their results in July 2026, while practical projects were presented during the event, with each presentation lasting about five minutes.
From Language Models to Models That Interact with the Physical World
The 2026 GENIAC cycle aims to expand on the program’s results since it began in February 2024, with an approach that extends beyond the Japanese market to applications with global reach. This includes developing models that can be used in industrial and service robots, rather than relying solely on models that understand text or images without performing actions in their surrounding environment.
Among the projects presented, SB Intuitions is working on a model for manipulating objects using two robotic arms, while Highlanders is developing a model for interacting with a human-like environment. The list of participants also includes MW, Telexistence, Muso Action, XNOVA, FastLabel, ugo, Preferred Networks, Preferred Robotics, and KDDI.
Preferred Networks presented the PLaMo-VL vision-language model, with 8 billion parameters, designed to give drones and robots spatial-perception capabilities based on Japanese images. Turing also presented a VLA model that combines vision, language, and action, with a focus on real-time control of autonomous vehicles and robots.
Data and Hardware Are Part of the Problem
The challenge highlighted by the program is not limited to designing a larger artificial-intelligence model. Robotic applications require suitable datasets, as well as sensors, motors, control units, and embedded devices capable of running models in real time. Accordingly, the data-ecosystem project includes organizations such as the University of Tokyo, Osaka University, FingerVision, AI Robot Research Institute (AIRoA), AIST, STATION Ai, and ELEMENTS.
The projects indicate that building data suitable for shared use and developing infrastructure for embodied AI will be crucial to transforming models from research demonstrations into components of actual robotic systems. LMM also presented a multimodal model for reading graphs and diagrams, verifying its performance using the Japanese JDocQA dataset.
Why Does This News Matter?
The actual change here is the redirection of government support from developing general models toward a more complex chain encompassing data, software, hardware, and integration with robots. This is becoming increasingly important because Japan’s Ministry of Economy, Trade and Industry aims, according to the article, to deploy 1,000 AI-powered robots across 18 fields by 2040.
However, the announcement does not yet prove that this goal has been achieved or that the models are ready for broad commercial production. The success of robots will also depend on data quality, operational efficiency on resource-constrained devices, the safety of interaction with humans, and developers’ ability to integrate models into reliable mechanical systems. GENIAC therefore appears closer to building a long-term industrial and research foundation than to launching a single market-ready product.