VLA-Arena Dataset (L1 - Medium Variant)
About VLA-Arena
VLA-Arena is an open-source benchmark designed for the systematic evaluation of Vision-Language-Action (VLA) models. It provides a complete and unified toolchain covering scene modeling, demonstration collection, model training, and evaluation. Featuring 150+ tasks across 11 specialized suites, VLA-Arena assesses models through hierarchical difficulty levels (L0-L2) to ensure comprehensive metrics for safety… See the full description on the dataset page: https://huggingface.co/datasets/VLA-Arena/VLA_Arena_L1_M_rlds.
