This repository contains the VLA-IT dataset, a curated 650K-sample Vision-Language-Action Instruction Tuning dataset, and the SimplerEnv-Instruct benchmark. These are presented in the paper InstructVLA: Vision-Language-Action Instruction Tuning from Understanding to Manipulation. The dataset is designed to enable robots to integrate multimodal reasoning with precise action generation, preserving the flexible reasoning of large vision-language models while delivering leading manipulation… See the full description on the dataset page:
https://huggingface.co/datasets/ShuaiYang03/VLA_Instruction_Tuning.
