MarcelTorneMIT

fold_pants_preferences

Real-robot trajectories for folding shorts on a Franka Panda with human pairwise preference labels across 45 judgment axes. Built for reward modeling and preference learning research with 536 episodes and 4,390 preference labels.

Downloads0
Episodes536
Hours~9.8

Why This Matters for Physical AI

This dataset enables research in reward modeling and reinforcement learning from human feedback (RLHF) by providing fine-grained pairwise preference judgments across multiple task-specific axes, advancing methods for learning complex manipulation behaviors from human preferences.

Technical Profile

Modalities
rgbproprioception
Robot Embodiments
Franka Panda
Action Space
joint_positions
Environment
lab
Task Types
manipulationfolding
Episodes
536
Total Hours
~9.8
Data Format
LeRobot
Annotation Types
preference_labelspairwise_comparisonsreward_labelstask_success_flags
License
MIT
Part of the fold_pants_preferences 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