CLARU
rtrtwrw2026apache-2.0

granular-material-pour -- policy rollouts

Policy evaluation rollouts for a granular material pouring task, with 197 audited runs analyzed in an accompanying paper and 230 total runs including post-cutoff data. Dataset includes imperfect real-world data with detailed scoring and error analysis.

Downloads50
Episodes230

Why This Matters for Physical AI

This dataset demonstrates the importance of handling imperfect real-world robotic data for policy evaluation, including measurement artifacts, protocol failures, and the transparency needed when reporting results with known errors in physical AI systems.

Technical Profile

Modalities
weight_sensorload_cell
Environment
lab
Task Types
pouringmanipulation
Episodes
230
Annotation Types
outcome_labelsmanual_noteserror_measurements
License
apache-2.0
Part of the granular-material-pour family

Access

View on HuggingFace

Need custom weight_sensor data?

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

Request a Sample Pack

Related Datasets

granular-material-pour -- flour, single target

122 teleoperated episodes of pouring flour onto a scale using a YAM arm, with the instruction to pour flour onto the scale at a nominal target of 300g. Dataset includes 2.80 hours of data at 50 Hz with 3 cameras and a 1 kg load cell.

rgbproprioceptionforce_torque
310 downloadsOct 2026apache-2.0

granular-material-pour -- buckwheat, exceeding pours

161 teleoperated pours of buckwheat onto a scale using a YAM arm, demonstrating overshoot correction behavior where the operator deliberately pours past the target and removes excess material. Dataset contains 2.27 hours of video from 4 cameras and load cell measurements at 30 Hz.

rgbproprioceptionforce_torque
93 downloadsOct 2026apache-2.0