StaticEmbodiedBench is a dataset for evaluating vision-language models on embodied intelligence tasks, as featured in the OpenCompass leaderboard.
It covers three key capabilities:
Macro Planning: Decomposing a complex task into a sequence of simpler subtasks.
Micro Perception: Performing concrete simple tasks such as spatial understanding and fine-grained perception.
Stage-wise Reasoning: Deciding the next action based on the agent’s current state and… See the full description on the dataset page:
https://huggingface.co/datasets/xiaojiahao/StaticEmbodiedBench.
