How we are building a set of quality metrics to keep improving data quality and the performance of final policies on challenging real-world problems.

OpenGVL: a benchmark for estimating task progress across diverse manipulation tasks, for both robotic and human embodiments.