Peer-reviewed papers, preprints, methodology notes, and reproducibility artifacts. All open access, all community-reviewable.
Frontier robotics models converge on the same bottleneck: narrow real-world data is expensive to capture but generalises poorly, while pure simulation is cheap but suffers from the sim-to-real gap. This paper introduces Real-to-Sim Enrichment (RSE), a systematic method for turning a single narrow real capture into a distribution of validated simulation variants, scored across a V0 → V3 taxonomy representing physical fidelity and behavioural transferability. We evaluate on three cleaning-domain tasks and two material-handling primitives.
