AGIBOT releases AGIBOT WORLD 2026 Theme 3

Shanghai / Stockholm, September 1, 2026, AGIBOT has released AGIBOT WORLD 2026 Theme 3: Reinforcement Learning as open source. The dataset contains real robot data from expert demonstrations.

11,430 real-world trajectories · 98,159 annotated sub-moments · successes, failures and human interventions Shanghai / Stockholm, September 1, 2026, AGIBOT has released AGIBOT WORLD 2026 Theme 3: Reinforcement Learning as open source. The dataset contains real-world robot data from expert demonstrations, autonomous policy runs and human corrections. The first release includes 11,430 real-world trajectories across 14 industrial and home-related tasks (1,024 successful and 1,369 failed runs). Detailed annotations cover task progress, completion, errors, disturbances and human interventions, precisely the structured feedback needed for robots to learn both to succeed and to detect problems and hand over control to a human when necessary. Why it matters to Nordic customers Large demonstration datasets teach robots the ideal way to perform a task. Real-world operations are more complex: robots succeed, fail, deviate or require human assistance. Theme 3 captures all of this – 98,159 annotated sub-moments, 26,493 disturbance segments, 5,795 error segments and 10,684 segments with human intervention. It’s the same intelligence layer that powers the platforms we deliver in the Nordics. It’s what allows our X2, D1 and A3 to identify errors in real time, flag risk and know when to hand control back to a human operator – whether on a factory floor, in a warehouse or at an event. Curious about how this translates into real-world operations? Explore X2, D1 or A3, or contact AI Robotics Sweden.