CLARU
projectsim2026cc-by-4.0

Real-to-Sim YAM Benchmark: run artefacts

A benchmark dataset containing 46 batches of simulated robot manipulation trials across multiple tasks (blocks, bottles, bowls, clear_table, latte) with real reference scores from Robocurve evaluations. Includes raw logs, results, and reproducible scoring tables for evaluating real-to-sim transfer in robot learning.

Downloads24
Episodes920

Why This Matters for Physical AI

This benchmark evaluates the effectiveness of asset reconstruction in real-to-sim robot learning, providing reproducible metrics for assessing sim-to-real transfer quality across multiple manipulation tasks and policies.

Technical Profile

Environment
simulation
Task Types
manipulationpick_and_placegrasping
Episodes
920
Data Format
JSONL
Annotation Types
reward_labelsaction_labels
License
cc-by-4.0
Part of the Real-to-Sim YAM Benchmark family

Access

View on HuggingFace

Need custom physical AI data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets

Real-to-Sim YAM Benchmark: authored assets

Sixteen sim-ready 3D object assets reconstructed from photographs for real-to-sim robot manipulation evaluation, with physics properties including mass, friction, and restitution for tabletop tasks.

3d_meshusdpbr_textures
113 downloadsOct 2026cc-by-4.0

Real-to-Sim YAM Benchmark: scene splats

Two Gaussian-splat scenes for real-to-sim robot evaluation, comparing authored and default reconstructions built from handheld phone video using 3D and 2D Gaussian Splatting techniques.

rgbgaussian_splattingpoint_cloud
20 downloadsOct 2026cc-by-4.0