Robot Company Releases Free Dataset to Speed Up AI Research

Teaching robots to navigate messy kitchens and bustling offices just got easier thanks to a Shanghai company sharing its hard-won data with the world.

AGIBOT today released AGIBOT WORLD 2026, a massive open-source dataset designed to help researchers train robots for real-world environments. Unlike lab experiments where everything is controlled and predictable, this data captures the chaos of actual homes, stores, and workplaces.

The company took an unusual approach to collecting information. Instead of having robots repeat the same scripted tasks over and over, human operators controlled the machines in real time, responding to whatever happened naturally. This means the data shows robots handling unexpected situations, like reaching around obstacles or adjusting when objects move.

What makes this dataset special is how complete it is. The robots didn't just record what they saw with cameras. They captured force feedback from touching objects, spatial data from sensors, and detailed information about how every joint moved. AGIBOT even built matching virtual environments so researchers can test ideas in simulation before trying them on real robots.

The first batch of data focuses on imitation learning, where robots watch demonstrations and learn to copy them. It includes hundreds of hours of footage from commercial spaces, complete with detailed labels explaining what's happening at each step. Crucially, the dataset even includes mistakes and corrections, showing robots how to recover when things go wrong.

The Ripple Effect

By making this data free and accessible, AGIBOT is lowering the barriers for robotics researchers everywhere. Small labs and startups that couldn't afford to collect their own massive datasets can now tap into the same resources as major companies. That means more minds working on the challenge of getting robots out of factories and into everyday life.

The release is just the first of five planned phases, each focusing on a different aspect of robot intelligence. Future batches will tackle other key challenges in getting machines to work safely and effectively alongside people.

AGIBOT joins a growing movement of companies sharing their data to accelerate progress across the entire field. When everyone has access to better training information, robots improve faster for everyone.

The company's long-term vision is clear: democratize robot development so these helpful machines can move from research labs into homes and workplaces sooner rather than later.