Announcement
DiverseUMI-100 open-sourced with 12,000-plus robotics tasks
The creators say it draws on a 100-hour sample of their larger UMI dataset and includes dense annotations by Argus.
TLDR
The dataset’s creators say they are open-sourcing DiverseUMI-100, a collection of 12,000-plus tasks from a 100-hour sample, with dense annotations by their Argus robotics pipeline. In an October 6 post, they said their international UMI data-collection operation had grown from five to 90 operators in eight weeks, building a dataset of more than 1 million unique tasks and capturing more than 45,000 new tasks daily.
Combined views
—
2 Sources, first seen ago
