SPARC (Spatial Annotations from Robot Demonstrations with Reliability Calibration) is a risk-aware framework for automatically labeling robot demonstrations with structured spatial annotations. It is designed to provide reliable spatial labels at scale.

The framework assigns each annotation a reliability calibration, which is intended to help downstream learning algorithms weigh the annotations appropriately. By making the labeling process risk-aware, SPARC aims to reduce the impact of noisy or uncertain labels.

The approach is described in a paper on arXiv. The abstract does not include experimental details or comparisons with other methods, so the practical performance of SPARC is not yet documented in the source.