Telexistence Joins NVIDIA Cosmos Coalition

Telexistence Inc. (Headquarters: Ōta, Tokyo; CEO: Jin Tokimaka; hereinafter "TX") announces its participation in the NVIDIA Cosmos Coalition, a global ecosystem for Physical AI led by NVIDIA. This coalition brings together world model developers, AI innovators, and leaders in Physical AI to share and utilize the NVIDIA Cosmos platform, which includes open world models, datasets, and frameworks. TX has been selected by NVIDIA as a representative Japanese Physical AI company among numerous robotics and Physical AI firms. With this partnership, TX will further accelerate the social implementation and commercial deployment of humanoids, as well as the development of core robotics foundation models.

▍ Background and Purpose of Participation

TX has been advancing the social and commercial implementation of Physical AI that integrates autonomous operation and remote control in real-world environments such as retail, logistics, and manufacturing. Through practical operations, TX has accumulated hands-on knowledge and data on manipulation (object handling) and full-body motion of multi-joint robots. Through previous collaborations with NVIDIA, TX has implemented and validated NVIDIA’s world action model "DreamZero" and the open robotics foundation model "NVIDIA Isaac GR00T." These ongoing evaluations aim to leverage NVIDIA’s advanced technologies for TX’s proprietary robotics foundation model development. With this formal participation, TX will now fully integrate these technologies into its Physical AI development, accelerating the practical deployment of humanoids at the implementation level.

▶ Watch: From Real-World Humanoid Data to Physical AI

▍ Why Development Acceleration Is Possible — Technical Background

The biggest constraint in deploying humanoids in society is the volume and diversity of real-world data required to train robotics foundation models. While model performance heavily depends on data scale, collecting data via physical robots involves limitations in time, cost, and safety. NVIDIA Cosmos, as a world foundation model, can generate large-scale, physically consistent scenes and motions, enabling policy learning and validation in simulation before real-world deployment. This overcomes data collection bottlenecks and significantly shortens development cycles.

Furthermore, TX has accumulated proprietary real-world data from actual store environments and other operational sites. By integrating Cosmos, TX can now combine its unique real data with synthetic data to perform high-precision, high-speed training—from pre-training to post-training (domain adaptation). Additionally, full-stack integration across Cosmos, Isaac GR00T, and simulation environments streamlines sim-to-real transfer. Lightweight post-training on edge GPUs (NVIDIA RTX PRO 6000 Blackwell) enables rapid adaptation and improvement for each deployment site, drastically reducing the time to commercial deployment of humanoids.

▍ Areas TX Will Advance Through This Participation

- Accelerated development of robotics foundation models by fusing proprietary operational data with synthetic data generated via Cosmos

- Social implementation and commercial deployment of general-purpose humanoids in retail, logistics, and manufacturing environments

- Contribution to the Cosmos platform as a data and validation infrastructure essential for Physical AI

▍ Future Outlook

Through this ecosystem, TX will lead the integration of Japan-originated humanoids and robotics foundation models, gradually expanding real-world deployments in actual stores and operational sites. In response to the growing social challenge of labor shortages, TX will drive solutions beyond R&D—focusing on implementation and operational-level impact.

Looking ahead, TX envisions a future where humanoids reach truly practical performance levels: operation speeds sufficient for real use, smoother motions, and generalization capabilities that maintain accuracy across changing environments. These remaining challenges are frontier topics pursued by top research institutions and companies worldwide. TX will leverage its strengths—large-scale real-world data from commercial environments like convenience stores, a full-stack system that handles everything from hardware to data collection, foundation models, and control, and a large-scale learning infrastructure powered by cutting-edge GPU clusters—to systematically solve these challenges and accelerate the social implementation of Physical AI.

Currently hiring for the following positions:

- Next-Generation Logistics Robotics Hardware Engineer

- Robotics AI Foundation Model Lead

- Humanoid Robotics Systems Engineering Lead

▶ Job Details and Applications: https://tx-inc.com/en/careers/

▍ About Telexistence

TX is a Physical AI company building foundational technologies for intelligent robots that operate in the real world. Guided by the vision "to grasp every single object in the world with our own 'hands'," TX develops and provides integrated systems combining AI such as robotics foundation models, custom-designed robot hardware, and operations. By co-designing and co-operating AI for perception and decision-making with robots for physical execution, TX achieves scalable social implementation of Physical AI in core industries.

▍ Contact for This Announcement

Telexistence Inc., Public Relations: [email protected]