maximilianofir2025
Sreetz vision-trainable 20k GR00T N1.7 model — MaxO RC validation
Evaluation of a vision-trainable GR00T N1.7 model on a vial-picking manipulation task, achieving 79% success rate (79/100 episodes) on the Newton-So101-Teleop-Vials-To-Rack-Eval-v0 benchmark.
Downloads0
Episodes100
Why This Matters for Physical AI
This evaluation dataset demonstrates the performance of a vision-based manipulation model on structured pick-and-place tasks, providing empirical validation of sim-to-real capable policies trained with domain randomization.
Technical Profile
- Modalities
- rgb
- Robot Embodiments
- Newton
- Action Space
- joint_targets
- Environment
- simulation
- Task Types
- pick_and_placemanipulation
- Episodes
- 100
- Annotation Types
- language_instructionsreward_labels
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack