I built Calibra a layer for robotics dataset quality and intelligent data curation
We are exploring how teams can automatically audit data quality, detect problematic demonstrations.Therefore you can build smaller and high-quality training sets before spending GPU time. So you can reduce your training costs up to %70.
Calibra helps robotics teams audit dataset integrity, measure quality and coverage, and build smaller training sets. v0.8 adds measured training experiments: record real GPU-hours, wall-clock time, energy, and policy performance, then compare Full vs Random vs Calibra across retention levels.
I've been building Calibra, an open-source dataset observability and coreset optimization tool for robot learning.
It checks things like timestamps, duplicate/frozen frames, blurry cameras, jerky motion, action-state divergence, quality, and behavioral coverage.
The interesting part is the optimization side: across several public robotics datasets, we've found that quality-aware selection can preserve model performance while using substantially less training data than the full dataset or random selection.
Calibra: Train with Less DataTrain robot policies with up to 75% less data
