Korea’s biggest manufacturers back Config, the TSMC of robot data

2026-05-11 · Source: Robotics News | TechCrunch · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Intermediate, short

Summary

Config, a Seoul- and San Jose-based startup specializing in data for robotics foundation models (RFMs), has secured $27 million in an oversubscribed seed round, bringing its total funding to $35 million at a valuation exceeding $200 million. Samsung Venture Investment led the round, with strategic participation from Hyundai Motor's ZER01NE Ventures, LG Technology Ventures, and SKT America. Founded in January 2025 by former Meta researcher Minjoon Seo, Config focuses on collecting and transforming human motion data for robot training, rather than building robots directly. The company has amassed over 100,000 hours of human motion data, significantly larger than comparable open-source datasets, and aims to scale to 1 million hours. Config's unique approach involves converting data to suit robot movements before model training, a core technical differentiator.

Key takeaway

For entrepreneurs developing robotics AI, your focus should be on specialized data solutions rather than just model development. The high cost and complexity of collecting and preparing physical training data make it a critical bottleneck. Partnering with or investing in data-centric platforms like Config can significantly reduce development costs and accelerate the deployment of more capable, proprietary robot AI systems.

Key insights

Asia's manufacturing strength is driving investment into physical AI, exemplified by Config's robotics data platform.

Principles

• Better data is key to more useful robots.
• Physical AI data collection is inherently costly.
• Data conversion is critical for robot-specific training.

Method

Config collects human motion data in studios and fields, then transforms it to be robot-native before model training, akin to language translation for optimal fit.

In practice

• Focus on data quality for robotics AI.
• Consider specialized data providers for RFM training.
• Explore data conversion techniques for robot-specific tasks.

Topics

• Config
• Robotics Foundation Models
• Robot Data Collection
• Human Motion Data
• Physical AI

Best for: Entrepreneur, Director of AI/ML, Investor, Robotics Engineer

Related on AIssential

• Collecting robot training data is dirty, unglamorous work. Some AI labs are already paying XDOF to do it. — Robotics AI development is bottlenecked by scarce physical interaction data, driving demand for specialized data collection and annotation infrastructure.
• [Korean Startup Weekly News #118] Physical AI Takes Off in Korea - Wowtale — Korean AI startups are shifting towards physical, vertical, and infrastructure-driven solutions with global ambitions.
• China is training a robot future — one folded shirt at a time — China is overcoming robotics data scarcity by locally crowdsourcing real-world task videos, leveraging labor costs and public support.
• Neura Launches Europe's Largest Physical AI Training Center — High-quality, realistic training data is the primary challenge for advancing intelligent robotics, not hardware.
• Before we spend months processing open-source robotics datasets, tell us why this is a bad idea [D] — The robotics field faces a data interoperability problem, not scarcity, hindering dataset reuse across diverse systems.
• Tech companies desperately want to film you doing chores — Physical AI development faces a critical data bottleneck due to the complexity and cost of acquiring real-world interaction footage.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Robotics News | TechCrunch.