MarcelTornemit

put_cube_in_bowl_preferences

Real-robot trajectories for a cube-in-bowl manipulation task on a Franka Panda with human pairwise preference labels across multiple judgment axes. Built for reward-model and preference-learning research with 540 episodes and 2,348 preference labels.

Downloads0
Episodes540
Hours1.2

Why This Matters for Physical AI

This dataset enables training reward models and preference-based learning algorithms through rich pairwise preference annotations on real robot manipulation, advancing methods for learning from human feedback in robotics.

Technical Profile

Modalities
rgbproprioception
Robot Embodiments
Franka Panda
Action Space
joint_positions
Environment
lab
Task Types
manipulationpick_and_place
Episodes
540
Total Hours
1.2
Data Format
LeRobot
Annotation Types
language_instructionsreward_labelspairwise_preferences
License
mit
Part of the put_cube_in_bowl_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