This project fine-tunes lerobot/smolvla_base using the HuggingFaceVLA/libero dataset to improve vision-language-action understanding for multi-task robotic manipulation.
Overview
Model: smolVLA
Dataset: LIBERO
Training: Supervised fine-tuning (SFT)
Task: Vision-Language-Action learning from demonstrations
Dataset
LIBERO provides multi-task robotic manipulation demonstrations with visual observations, language instructions, and action sequences.
Training
The model was fine-tuned on the LIBERO dataset following the original smolVLA instructions, mapping visual observations and language instructions to robot actions.
