BookHou/smolvla-rgbd-frozen-expert-b32-7500

# SmolVLA RGB-D Frozen Expert B32 7500

SmolVLA policy trained for an SO-101 flower-pot manipulation task using three visual inputs, with the vision encoder frozen.

## Training configuration

• —Training steps: 7,500
• —Batch size: 32
• —Learning rate: 1e-4
• —Vision encoder frozen: Yes
• —Expert-only training: Yes
• —Visual inputs: 3
• —Robot state dimension: 6
• —Action dimension: 6
• —Action chunk size: 50
• —Base VLM: HuggingFaceTB/SmolVLM2-500M-Video-Instruct

## Dataset

The model was trained on a 200-episode RGB-D robot manipulation dataset.

## Intended use

This model is preserved as a frozen-vision baseline for comparison with the fully fine-tuned 9,280-step model.

## Limitations

The policy is specific to the original robot, camera arrangement, workspace, and task. Recalibration and supervised safety testing are required before physical deployment.