Generative AI Speeds Up Robot Development: SO-101 Demo Showcases Japan’s Physical AI Potential at IVS2026

September 7, 2026

futureofwork innovation japan generativeai japannews japanlifestyle opensource japanbusiness artificialintelligence startupjapan japanvisa visa expatjapan livinginjapan japanstartup techinjapan robotics claude claudecode physicalai ivs2026 japanesetechnology robotarm aiinnovation amazonbedrock aws awsstartups classmethod hugface lerobot opensourcerobotics coding vibecoding foreigninjapan

Summary

At IVS2026, Classmethod demonstrated an SO-101 open-source robot arm sorting wooden blocks through preset commands. The prototype was reportedly created in about two weeks by a developer without prior robotics experience. Claude Code supported coding, technical research, testing and the development of the robot’s control interface. The project used existing open-source hardware and software rather than building the entire robot from scratch. The demonstration also showed why physical AI is challenging, as lighting, camera angles and object placement can affect performance. Classmethod highlighted AWS services, Claude through Amazon Bedrock and support for AI-driven development. Eligible startups may also be able to access up to 1,000 dollars in AWS credits, subject to conditions. Read the full article to see what this experiment could mean for Japan’s robotics and startup ecosystem.

Open-source robotics meets AI-assisted development

A robot arm sorting wooden blocks may look like a simple demonstration, but the SO-101 display at IVS2026 offered a glimpse into how generative artificial intelligence could change the way physical machines are developed. At the AWS booth, Japanese technology company Classmethod demonstrated a robot arm that rearranged blocks marked with letters such as “I,” “V,” “S,” “A” and “W” in response to commands selected on a computer screen.

The demonstration was built by combining the open-source SO-101 robotic arm with Claude Code, Anthropic’s AI coding agent. According to the explanation provided at the event, the team created the exhibition prototype in approximately two weeks, despite the developer involved having no previous experience in robotics or mechanical control development.

Inside the SO-101 demonstration

The SO-101 is the successor to the SO-100, developed jointly by RobotStudio and Hugging Face. Its main structural components can be produced using 3D printers and combined with commercially available servo motors and other parts. The platform is also compatible with LeRobot, Hugging Face’s open-source framework for training and developing robots.

Under the message “From Data to Physical AI - Build with AWS,” the booth brought together the robot arm, a camera, edge-computing equipment and AWS cloud services. A controller application displayed buttons including “AWS,” “IVS,” “Custom,” “Wave,” “Break” and “STOP.” When a visitor or staff member selected a button, the arm performed a predefined task, moving the wooden blocks into a specified sequence.

The exhibit also highlighted an important difference between software-only applications and physical AI systems. A program running entirely on a computer can often be tested under controlled digital conditions. A robot operating in the real world must deal with changes in lighting, camera angle, object placement, equipment positioning and hardware characteristics. Even small physical differences can cause a task to fail.

How Claude Code supported the project

Claude Code operates through a terminal and assists developers with writing and modifying code, researching technical information, running commands and testing implementations based on natural-language instructions. The development team reportedly interacted with the tool while working on the robot’s control functions and user interface.

This approach is sometimes described as “vibe coding,” in which developers use conversational instructions to guide an AI system through software development. However, the two-week timeline should be understood accurately. The team did not create the robot’s mechanical structure or all of its underlying software from scratch. Instead, it used the existing SO-101 platform and open-source software assets, then built exhibition-specific controls and interfaces on top of them.

Even with that qualification, the project demonstrates how open-source robotics and AI coding assistants may lower the barrier to experimentation. For Japanese startups, research teams and manufacturers, such tools could make it easier to test ideas before investing in larger-scale engineering projects.

AWS support and a more accessible developer ecosystem

Classmethod’s booth also introduced support for startups seeking to use AWS services. The company presented information about up to 1,000 U.S. dollars in AWS credits for eligible startups founded within the previous 10 years. The credit is not automatically granted to every company meeting that age requirement, and applicants must confirm the applicable services and conditions with Classmethod.

On March 2, 2026, Classmethod announced that it had signed a reseller agreement with Anthropic under the “Anthropic Authorized Reseller Program for Amazon Bedrock.” The arrangement allows the company to resell Claude delivered through Amazon Bedrock and support integration with existing AWS environments, including combined billing arrangements.

From exhibition demo to business application

Beyond providing Claude through Amazon Bedrock, Classmethod offers services covering AI-driven development processes, workflow optimization, system integration, operational support, cost management and internal training. The SO-101 demonstration served as a practical example of how those capabilities can be applied to a physical prototype.

For international visitors interested in Japan’s technology sector, the display reflected a broader trend: Japanese companies are increasingly combining established engineering expertise, open-source platforms and cloud-based AI tools. The road from a small booth demonstration to a reliable industrial robot remains substantial, but projects such as this show how generative AI is making early-stage experimentation faster and more approachable.