Open Models Almanac
Methodsdata curationdata mixture & curriculum

single mention · data curation

Pre-training on diverse robot-manipulation data

Pre-training on diverse robot-manipulation data is the first stage of a training framework for robot foundation models.1 It trains on a large mixture of manipulation data from many robot configurations and tasks, after which the model is evaluated out-of-the-box or fine-tuned on downstream tasks.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. a framework for training a robot foundation model, which we refer to as , that consists of pre-training on highly diverse data — π0\pi_{0}: A Vision-Language-Action Flow Model for General Robot Control Thanks: Physical Intelligence, San Francisco, California, USA.…
  2. pre-training on highly diverse data, followed by either out-of-box evaluation or fine-tuning to complex downstream tasks — π0\pi_{0}: A Vision-Language-Action Flow Model for General Robot Control Thanks: Physical Intelligence, San Francisco, California, USA.…

Also called pre-training on highly diverse data.

Filed under data curation › data mixture & curriculum