MarcelTorneMIT

plate_toast_preferences

Real-robot trajectories for a Franka Panda performing "put the toast in the plate" with human pairwise preference labels across 43 judgment axes. Built for reward-model and preference-learning research with 271 episodes and 2,671 preference labels.

Downloads0
Episodes271
Hours~1.5

Why This Matters for Physical AI

This dataset enables training of reward models through pairwise preference learning on real robot manipulation trajectories, advancing RLHF and preference-learning methods for robotic control.

Technical Profile

Modalities
rgbproprioception
Robot Embodiments
Franka Panda
Action Space
joint_positions
Environment
lab
Task Types
manipulationpick_and_place
Episodes
271
Total Hours
~1.5
Data Format
LeRobot
Annotation Types
preference_labelspairwise_comparisonsreward_labels
License
MIT
Part of the MarcelTorne preferences dataset collection 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