AGIBOT WORLD 2026 Theme 3

AGIBOT WORLD 2026 Theme 3 is an open-source dataset released by AGIBOT on September 1, 2026 as part of the AGIBOT WORLD 2026 initiative. It focuses on real-world robot reinforcement learning and is built from real-world robot experience data rather than simulation. The release brings together three complementary types of real-world trajectories. Expert demonstration trajectories record reference executions performed by human operators in real-world environments. Policy rollout trajectories show how models perform when executing independently. Human-in-the-loop correction trajectories capture the model's behavior before intervention and the point at which a human operator takes control. Together they form an end-to-end data pipeline spanning demonstration, autonomous execution and correction. First release figures, per AGIBOT's official announcement: the first release includes 11,430 real-world trajectories across 14 industrial and household tasks, including 1,024 successful policy rollouts and 1,369 failed policy rollouts. AGIBOT states that the first batch also contains 98,159 annotated subtask intervals, each with a completion-status label, together with 26,493 disturbance segments. Annotations cover task progress, completion, errors, disturbances and interventions. AGIBOT states that these annotations allow researchers to analyze execution at the process level rather than relying only on trajectory-level success or failure. Theme 3 extends the AGIBOT WORLD 2026 initiative from learning primarily from demonstration toward reinforcement learning, and AGIBOT states that it plans to continue expanding the initiative with additional datasets, benchmarks and research resources.