The SigmaKit is aimed at industries such as electronics, batteries, and food and beverage. | Source: CarbonSix
CarbonSix Inc., a developer of physical AI for manufacturing, yesterday launched a toolkit based on robot imitation learning. It said Sigmakit can be deployed directly onto the factory floor, allowing manufacturers to implement AI-powered robots without specialized expertise or training.
The toolkit uses imitation learning to eliminate the burdensome process of system configuration and constant adjustments traditionally required by human operators, said the Wilmington, Del.-based company.
“Sigma Kit is the first commercially available industrial robot AI solution based on imitation learning, capable of handling non-standardized and delicate tasks such as film attachment/removal, assembly, machine tending, cable fastening, and hanging operations,” stated Jehyuk Kim, co-CEO of CarbonSix. He led the development of SigmaKit.
“It can be applied across a wide range of industries, including mobile, consumer electronics, automotive components, food, and materials,” he added. “Since its launch, we have already received sales inquiries and reservations, and we are currently conducting PoC [proof-of-concept] projects with major global manufacturers.”
Imitation learning allows AI systems to learn directly from human demonstrations, enabling robots to reproduce complex actions and adapt flexibly to variable tasks, explained CarbonSix. Unlike traditional programming approaches, it makes automation possible for unstructured and non-routine processes once considered beyond the reach of robotics.
CarbonSix hopes to break down barriers to automation
The manufacturing sector has long faced barriers to full automation because frequent product changes and high variability in production tasks, CarbonSix said. Unlike humans, conventional robots have struggled to adapt flexibly to unstructured environments.
SigmaKit combines hardware and software to address these challenges, featuring:
- AI algorithms tailored to manufacturing processes
- Precision robotic grippers for delicate manipulation
- A teaching tool for simplified operation
- Sensor modules for adaptive recognition
With the kit, users demonstrate the target task repeatedly, building a training dataset. Depending on the task, a model can be generated in less than a day, CarbonSix claimed. The kit allows users to filter and review the collected data, making training more efficient.
Once it has gathered enough image and motion data, the system produces a robotic motion intelligence, or skill, that enables the robot to execute complex tasks. The skill can then be deployed directly to unstructured tasks, the company said.
CarboxSix was founded in 2024 and is led by co-CEOs Jehyuk Kim and Terry Moon. Moon previously served as chief strategy officer and vice president at SUALAB, a South Korean AI company acquired by Cognex Corp. in 2019 for approximately $260 million.
Kim is an AI and robotics researcher with academic experience from MIT, Yale University, Seoul National University, and KAIST. Dr. Hyungju Suh, the chief technology officer, is a robotic systems engineer. Suh received his Ph.D. from MIT CSAIL and participated in projects with research institutions such as the Boston Dynamics Robotics and AI Institute, Toyota Research Institute, and NASA Jet Propulsion Laboratory.
Editor’s note: RoboBusiness 2025, which will be on Oct. 15 and 16 in Santa Clara, Calif., will include tracks on physical AI and humanoid robots. Registration is now open.
Tell Us What You Think!