Prox Industries Selected for 'Physical AI Development Support Program by AWS Japan'
Prox Industries has been selected for the 'Physical AI Development Support Program by AWS Japan.' Through this initiative, the company aims to build a Real2Sim2Real VLA (Vision-Language-Action) learning pipeline to streamline the application of Physical AI to specific industrial use cases.
📋 Article Processing Timeline
- 📰 Published: March 31, 2026 at 20:11
- 🔍 Collected: April 1, 2026 at 13:39 (17h 27m after Published)
Prox Industries, Inc. (hereinafter "Prox Industries") is pleased to announce that it has been selected for the "Physical AI Development Support Program by AWS Japan," provided by Amazon Web Services Japan G.K. (hereinafter "AWS Japan").
Through this selection, our company will promote the construction of a Real2Sim2Real VLA (Vision-Language-Action) learning pipeline applicable to various environments, tasks, and hardware, aiming to streamline the process of applying VLA to specific use cases.
### Background
In recent years, the advancement of robot foundation models, including VLA, has significantly expanded the possibilities for robot control that integrates language, vision, and action. On the other hand, current VLA models tend to be highly dependent on the environments, tasks, and hardware assumed during training, often failing to demonstrate sufficient performance in out-of-distribution (OOD) domains that deviate from those conditions.
Furthermore, absorbing such diverse conditional differences into a single, general-purpose model remains highly challenging even at the forefront of current research, and in practical operations, there are many situations where fine-tuning for specific tasks is necessary. To deploy Physical AI into the real world, it is crucial to establish a cycle of learning, evaluation, and improvement tailored to individual use cases with high reproducibility and efficiency.
### Initiatives in this Program
Leveraging our strengths in simulation and manipulation technologies, Prox Industries will work on building a Real2Sim2Real VLA learning pipeline that can be applied to various environmental, task, and hardware conditions.
We will standardize the cycle of collecting demonstration data from real machines, reflecting it in a simulation environment (NVIDIA Omniverse), generating synthetic data, fine-tuning the VLA, and verifying it on real machines, establishing this as a general-purpose pipeline. By doing so, we aim to streamline the development and improvement processes when applying VLA to specific industrial use cases and build a more practical foundation for the introduction of Physical AI.
### Overview of 'Physical AI Development Support Program by AWS Japan'
This is a support program provided by AWS Japan for companies and organizations with a legal entity or base in Japan that are developing robot foundation models, such as Vision-Language-Action (VLA), on AWS. It aims to promote the use of AI in robotics by supporting the construction of a series of development pipelines, from data collection and preprocessing to model training, simulation, and deployment to real-world environments.
Selected companies and organizations will receive technical support from Physical AI specialists, AWS credits to cover a portion of development costs, opportunities for robotics and generative AI community building, and go-to-market support. The provision of AWS credits is intended for workloads related to robot foundation model development (including fine-tuning, simulation, synthetic data generation, etc.), with a total of up to 6 million US dollars planned for the entire program.
### About Prox Industries
Prox Industries is a startup originating from the Matsuo Lab at the University of Tokyo. With the mission of "Implementing Physical Intelligence from Japan," we aim to implement intelligence in the realm of physical labor—which has previously only been performed by humans—through the fusion of AI and robotics.
Gathering researchers and engineers specializing in core Physical AI technologies, we are leading multiple Physical AI projects, primarily for domestic enterprise companies. We are tackling advanced technical challenges, including Sim2Real and Real2Sim, while utilizing diverse real machines such as humanoids, quadruped robots, collaborative robots, and mobile manipulators, as well as simulation environments centered on NVIDIA Omniverse.
By connecting cutting-edge academic technology with the needs of industrial sites, we provide end-to-end support from R&D of robot foundation models and construction of learning pipelines to evaluation and improvement for real-machine deployment. In doing so, we are advancing the research, development, and implementation of Physical AI and working to build the next industrial foundation for Japan.
Company Overview
- Company Name: Prox Industries, Inc.
- Location: Hongo K&K Building 4F, 3-30-10 Hongo, Bunkyo-ku, Tokyo
- Representative: Yuji Kamei, Representative Director
- Company Website: https://prox-industries.jp/
Through this selection, our company will promote the construction of a Real2Sim2Real VLA (Vision-Language-Action) learning pipeline applicable to various environments, tasks, and hardware, aiming to streamline the process of applying VLA to specific use cases.
### Background
In recent years, the advancement of robot foundation models, including VLA, has significantly expanded the possibilities for robot control that integrates language, vision, and action. On the other hand, current VLA models tend to be highly dependent on the environments, tasks, and hardware assumed during training, often failing to demonstrate sufficient performance in out-of-distribution (OOD) domains that deviate from those conditions.
Furthermore, absorbing such diverse conditional differences into a single, general-purpose model remains highly challenging even at the forefront of current research, and in practical operations, there are many situations where fine-tuning for specific tasks is necessary. To deploy Physical AI into the real world, it is crucial to establish a cycle of learning, evaluation, and improvement tailored to individual use cases with high reproducibility and efficiency.
### Initiatives in this Program
Leveraging our strengths in simulation and manipulation technologies, Prox Industries will work on building a Real2Sim2Real VLA learning pipeline that can be applied to various environmental, task, and hardware conditions.
We will standardize the cycle of collecting demonstration data from real machines, reflecting it in a simulation environment (NVIDIA Omniverse), generating synthetic data, fine-tuning the VLA, and verifying it on real machines, establishing this as a general-purpose pipeline. By doing so, we aim to streamline the development and improvement processes when applying VLA to specific industrial use cases and build a more practical foundation for the introduction of Physical AI.
### Overview of 'Physical AI Development Support Program by AWS Japan'
This is a support program provided by AWS Japan for companies and organizations with a legal entity or base in Japan that are developing robot foundation models, such as Vision-Language-Action (VLA), on AWS. It aims to promote the use of AI in robotics by supporting the construction of a series of development pipelines, from data collection and preprocessing to model training, simulation, and deployment to real-world environments.
Selected companies and organizations will receive technical support from Physical AI specialists, AWS credits to cover a portion of development costs, opportunities for robotics and generative AI community building, and go-to-market support. The provision of AWS credits is intended for workloads related to robot foundation model development (including fine-tuning, simulation, synthetic data generation, etc.), with a total of up to 6 million US dollars planned for the entire program.
### About Prox Industries
Prox Industries is a startup originating from the Matsuo Lab at the University of Tokyo. With the mission of "Implementing Physical Intelligence from Japan," we aim to implement intelligence in the realm of physical labor—which has previously only been performed by humans—through the fusion of AI and robotics.
Gathering researchers and engineers specializing in core Physical AI technologies, we are leading multiple Physical AI projects, primarily for domestic enterprise companies. We are tackling advanced technical challenges, including Sim2Real and Real2Sim, while utilizing diverse real machines such as humanoids, quadruped robots, collaborative robots, and mobile manipulators, as well as simulation environments centered on NVIDIA Omniverse.
By connecting cutting-edge academic technology with the needs of industrial sites, we provide end-to-end support from R&D of robot foundation models and construction of learning pipelines to evaluation and improvement for real-machine deployment. In doing so, we are advancing the research, development, and implementation of Physical AI and working to build the next industrial foundation for Japan.
Company Overview
- Company Name: Prox Industries, Inc.
- Location: Hongo K&K Building 4F, 3-30-10 Hongo, Bunkyo-ku, Tokyo
- Representative: Yuji Kamei, Representative Director
- Company Website: https://prox-industries.jp/