galbenecho

Echo Foundry — VLA Dataset

10,133 episodes (60 hours) of physics-verified UR5e robot manipulation with reasoning annotations, ground-truth object poses, metric depth, and rejected planner alternatives for every frame.

Downloads4
Episodes10,133
Hours60

Why This Matters for Physical AI

This dataset provides the first large-scale corpus of robot manipulation with explicit reasoning traces, rejection margins, and ground-truth world state, enabling training of embodied VLMs that can explain and learn from both successes and failures in a sim-to-real transferable format.

Technical Profile

Modalities
rgbdepthproprioceptionlanguageforce_torque
Robot Embodiments
UR5e
Action Space
joint_positions
Environment
labsimulation
Task Types
manipulationpick_and_placestackinggrasping
Episodes
10,133
Total Hours
60
Data Format
LeRobot
Annotation Types
language_instructionsreward_labelsaction_labelsbounding_boxessegmentation
Part of the Echo Foundry 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