📊 Key Data
  • 1 million hours of human physical data collected, targeting 10 billion hours by 2030.
🎯 Expert Consensus

Experts agree that Maniformer's data infrastructure is pivotal for advancing physical AI, but its success hinges on navigating competitive tensions and global regulatory challenges.

about 14 hours ago
The Picks and Shovels of Physical AI: Inside Maniformer's Data Empire

The Picks and Shovels of Physical AI: Inside Maniformer's Data Empire

AUSTIN, Texas – November 16, 2026 — The most valuable asset in the next industrial revolution isn't a proprietary servo motor or a cutting-edge titanium chassis. It is the mundane, meticulously recorded data of a human hand folding a shirt, tightening a bolt, or navigating a cluttered warehouse aisle.

At the 10th Conference on Robot Learning (CoRL 2026) in Austin this week, the industry's focus notably shifted from the flashy mechanics of humanoid robots to the invisible infrastructure required to make them actually useful. While robotics hardware has advanced at a blistering pace, the foundation models that serve as the "brains" of these machines are starving for real-world interaction data. Unlike large language models, which can passively scrape billions of text documents from the internet, physical AI requires multimodal, physics-bound data that must be painstakingly collected in the real world.

Enter Maniformer, a Shanghai-based robotics data company that emerged from stealth earlier this year and is rapidly positioning itself as the premier infrastructure provider for the embodied AI boom. Co-hosting a high-profile Research Social Night with its parent company AGIBOT, the data provider drew over 60 leading researchers from Meta, NVIDIA, Google DeepMind, and top-tier universities. The gathering underscored a fundamental industry pivot: the race to build general-purpose robots is no longer just a hardware challenge; it is a massive, distributed data logistics operation.

Breaking the Robotics Data Bottleneck

To understand the bottleneck holding back physical AI, one must look at how robots traditionally learn. Historically, data collection has been robot-centric—researchers painfully teleoperate a $100,000 robotic arm to perform a single task hundreds of times. It is a slow, unscalable, and prohibitively expensive process.

Maniformer's strategic breakthrough is its "embodiment-free" approach. Instead of using robots to collect data for robots, the company is using humans. Through its MEgo series of capture hardware, human operators wear sensor-laden gear to record their daily tasks. The MEgo View, a dual-perspective headset combining a 300-degree panoramic fisheye camera with a wrist-mounted close-up lens, captures high-fidelity visual streams. Paired with the MEgo Gripper—a cable-free device that records millimeter-precision spatial positioning and tactile force—the system captures the exact visual, motion, and interaction feedback of a human performing a physical task.

"The realization is that human video provides the high-level reasoning, but real-world tactile interaction provides the essential action feedback," noted one robotics researcher attending the CoRL after-party. "If you want a robot to recover from a failure gracefully, you need thousands of hours of data showing how a human recovers from that exact same failure."

To scale this, the data provider launched ManiGo—known domestically in China as Mifengpai—a crowdsourcing platform that essentially turns ordinary citizens into paid "robot trainers." Users can rent MEgo capture gear and receive compensation for recording themselves performing tasks across 22 categories, from retail stocking to home dishwashing. By transforming everyday human labor into structured data assets via its MEgo Engine processing pipeline, the company has already surpassed one million hours of human physical data, with an ambitious target of 10 billion hours by 2030.

The 'Picks and Shovels' Supply Chain

From a business intelligence perspective, Maniformer represents the maturation of the humanoid robotics supply chain. In any gold rush, the most reliable returns often go to those selling the picks and shovels. Rather than competing directly in the crowded humanoid hardware market, this enterprise is building the ancillary infrastructure that every hardware manufacturer will eventually need.

This infrastructure play has not gone unnoticed by institutional capital. Founded just in February 2026, the company quickly secured hundreds of millions of yuan in seed and angel financing led by Sequoia Capital China, Baidu Ventures, and later, China Telecom Group. This war chest is being deployed to build a comprehensive data ecosystem, which includes co-founding RoboColiseum—an embodied AI evaluation platform—alongside industry heavyweights Hugging Face, NVIDIA, and Tsinghua University.

However, this strategic positioning is not without its complexities. Maniformer is a subsidiary of AGIBOT, a prominent manufacturer of humanoid and quadruped robots. Yao Maoqing serves as both the Chairman and CEO of the data company and as a senior vice president at AGIBOT. This dual role creates a powerful "leadership loop" that tightly integrates data collection with model training and field deployment.

Yet, it also introduces a delicate market dynamic. "If you are a competing robotics OEM, you might think twice before routing your proprietary task data through an infrastructure provider so closely tied to your direct competitor," explained one industry analyst tracking the space. Navigating this tension—proving it can serve as a neutral, secure data pipeline for the broader industry while maintaining its strategic ties to AGIBOT—will be the company's most significant corporate hurdle.

The Geopolitics of Embodied Data Nodes

The implications of this data race extend far beyond the laboratory. As embodied AI models become integral to national manufacturing, logistics, and defense capabilities, the data that trains them is becoming a critical geopolitical asset.

Maniformer is aggressively expanding its global footprint. While it currently operates across more than 20 innovation centers in China, it has already established international network nodes in over five countries. Concurrently with CoRL in Texas, the company exhibited at NexTech Week 2026 Autumn in Tokyo, targeting high-value industrial and service data in the Japanese market.

This transnational data pipeline introduces a labyrinth of regulatory challenges. Capturing high-fidelity video, spatial mapping, and human behavioral data across borders means navigating an alphabet soup of privacy frameworks, from China's Personal Information Protection Law (PIPL) to Europe's GDPR and various US state privacy acts. Furthermore, the crowdsourcing model requires rigorous worker safety protocols and insurance frameworks, especially when deploying ordinary citizens to capture data in active industrial or warehousing environments.

As the intelligent agent era accelerates, the companies that can successfully manage these operational, regulatory, and geopolitical hurdles will dictate the pace of the next industrial revolution. The discussions at CoRL 2026, featuring leading minds from Harvard, UCLA, and Northwestern, made one thing abundantly clear: the hardware is ready, the algorithms are waiting, but the ultimate winner of the physical AI race will be the entity that controls the highest quality real-world data.

Topics & Related

Event:
Seed Round
Product Launch
Theme:
Artificial Intelligence
Sector:
Robotics & Automation
Data & Analytics

📝 This article is still being updated

Are you a relevant expert who could contribute your opinion or insights to this article? We'd love to hear from you. We will give you full credit for your contribution.

Contribute Your Expertise →
UAID: 55071