sii-rhos-ai2025MIT

ViFailback Dataset — LeRobot

Real-world ALOHA dual-arm manipulation trajectories designed for studying failure diagnosis, localization, corrective guidance, recovery, and learning from failed demonstrations. Contains 5,202 episodes with frame-level failure annotations and keyframe labels.

Downloads0
Episodes5202
Hours14.41
Likes1

Why This Matters for Physical AI

Provides large-scale real-world manipulation data with explicit failure annotations and recovery guidance, enabling research into failure diagnosis and learning from negative demonstrations for robotic manipulation.

Technical Profile

Modalities
rgbproprioceptionjoint_positionsjoint_velocitiesjoint_effort
Robot Embodiments
ALOHA
Action Space
joint_positions
Environment
lab
Task Types
manipulationfailure_recoverygrasping
Episodes
5202
Total Hours
14.41
Data Format
LeRobot
Annotation Types
language_instructionsfailure_labelskeyframe_labelsfailure_type_labels
License
MIT
Part of the LeRobot family

Access

Need custom rgb data?

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

Request a Sample Pack

Related Datasets