MangoGoes
LIBERO 4in1 Wan2.2-VAE Latent Cache
Pre-encoded latent tensors for LIBERO 4 suites (libero_spatial, libero_object, libero_goal, libero_10) using Cosmos Wan2.2-VAE, enabling 5-10× training speedup by skipping on-the-fly VAE encoding during action-policy training.
Downloads16
Episodes1717
Why This Matters for Physical AI
This latent cache dataset accelerates training of vision-language-action models and imitation learning policies by pre-computing expensive VAE encodings, enabling faster iteration on embodied AI research across diverse manipulation tasks.
Technical Profile
- Modalities
- rgbproprioception
- Robot Embodiments
- Franka Panda
- Action Space
- joint_positions
- Environment
- lab
- Task Types
- manipulationpick_and_placegrasping
- Episodes
- 1717
- Data Format
- PyTorch
- Annotation Types
- action_labels
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack