FactoryNet4MIT

FactoryNet

A multi-embodiment industrial time-series corpus containing 56,591 task executions (14,847 real, 41,744 simulated) across 7 robot embodiments with 113M logged timesteps, structured in a unified Setpoint-Effort-Feedback-Context schema for anomaly detection and predictive maintenance.

Downloads37
Episodes56,591

Why This Matters for Physical AI

FactoryNet enables sim-to-real transfer and cross-embodiment anomaly detection by providing a unified control-theoretic schema (S-E-F-C) that standardizes sensor and command data across diverse industrial robots, addressing the critical challenge of learning generalizable dynamics models across different platforms.

Technical Profile

Modalities
time-seriesjoint_positionsjoint_velocitiesjoint_torquemotor_currenttcp_poseaccelerometercontact_force
Robot Embodiments
UR3UR5UR10UR30KUKA KR10CNC gantry
Action Space
joint_positions
Environment
labsimulation
Task Types
pick_and_placescrewingpeg_insertionmachining
Episodes
56,591
Data Format
parquet
Annotation Types
anomaly_labelsfault_labelstask_phasecounterfactual_pairs
License
MIT
Part of the FactoryNet family

Access

Need custom time-series data?

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

Request a Sample Pack

Related Datasets