Insider Brief
- Maxinsights raised $30M in seed funding led by Peak XV and launched MaxLoop and MaxGlove to improve the collection, quality control and processing of real-world training data for physical AI systems.
- The company said it has captured more than 2 million hours of data, including 1.5 million model-ready hours, using more than 5,000 contractors across six countries to record real-world tasks.
- MaxLoop scores and extracts information from footage as it arrives, while MaxGlove is being developed with Genesis AI to capture touch and force data, with Maxinsights targeting 1 million hours of force data in 2027.
Maxinsights has raised $30 million in seed funding and launched two products designed to improve the collection and processing of real-world training data for physical AI systems.
According to Maxinsights, Peak XV led the round, with participation from Tola Capital, South Park Commons, INCE Capital and Alumni Ventures.
Maxinsights collects first-person footage of people carrying out everyday work in homes, restaurants, hotels and other real-world settings. The company said it has captured more than 2 million hours of data, including 1.5 million hours that are model-ready and available for delivery.
“We built the whole pipeline ourselves, from the hand-tracking models to the platform that scores every operator’s footage as it arrives, because nobody was going to sell us the pieces that make messy real-world data usable,” founder and CEO Yuguang (YG) Yuan said in the annnouncement. “Robots get deployed into real kitchens, not staged ones, and something unexpected happens every few minutes. Edge cases are the job, and the only way to get them is to record real work in real homes. Collecting one great hour of data is easy, and holding quality across millions of hours and thousands of people is what we built the company to do.”
More than 5,000 contractors across six countries use Maxinsights camera hardware while performing tasks such as loading dishwashers, preparing food and washing cars. The company hires and manages workers through staffing firms rather than outsourcing collection to third-party data vendors.
The company said all 200,000 hours of egocentric data used to train Genesis AI’s GENE-26.5 model came from Maxinsights. It also described itself as the largest data provider for Dyna Robotics’ Dyna-2 World-Action Model.
Building the Data Pipeline
Maxinsights handles collection, quality control, annotation, curation and delivery within its own platform.
Footage is first checked through automated and human review for issues including image quality, task completion, visibility, synchronization and metadata. Proprietary models then label information about the environment, task, subtasks and instructions, along with hand poses, body movement and interactions with objects.
Customers can search and rebalance datasets by factors including task, location, behavior and complexity before receiving raw or processed data. Feedback from customers and models can then be incorporated into subsequent collection.
MaxLoop and MaxGlove
The company also introduced MaxLoop, software that evaluates and corrects data collection as it happens, and MaxGlove, a sensor glove being developed with Genesis AI to capture touch and force information that cannot be derived from video alone.
The two new products extend that data pipeline into real-time quality control and physical interaction data:
- MaxLoop: Scores footage as it arrives and gives operators feedback the same day, compared with one to two weeks previously. Maxinsights said the system has reduced review costs by about 80% and can process about 450,000 hours of footage a month. Its models also extract information including hand pose, object state, contact and force from individual clips.
- MaxGlove: Captures grip force, pressure, resistance, slipping and other physical signals that video alone cannot provide. Developed with Genesis AI, the glove is designed to work with Maxinsights’ existing egocentric data-collection system and will be deployed through its contractor network. Maxinsights said existing sensor gloves can require lengthy calibration and may provide only 10 to 20 minutes of usable capture a day. The company is targeting 1 million hours of force data in 2027.
Maxinsights said force and torque data could help robot models determine how firmly to grasp objects, recognize secure contact and adjust when objects shift or slip.
The Company
The company was founded by Yuan and CTO Kun Wang. Yuan previously co-founded industrial robotics company Mega Gear Robotics, while Wang previously worked at Google X and Waymo and later co-founded autonomous-driving company QCraft. Maxinsights noted its engineering team includes people with experience at Google, Microsoft, Meta, Waymo and Hulu.
According to Maxinsights, which is based in Santa Clara, Calif. with offices in China and Indonesia, it serves more than 30 customers, including AI labs and foundation-model developers, with eight customers on recurring contracts. Offices in Turkey and Mexico are planned as well, the company said.
