MECoBench is a benchmark for systematically evaluating multimodal multi-agent collaboration in embodied environments.
It contains 192 task cases constructed in VirtualHome, covering two collaboration structures:
Parallel collaboration: 96 task cases in which agents operate in a
shared environment and can complete independent subtasks concurrently.
Sequential collaboration: 96 task cases in which agents operate in
disjoint spatial regions and must coordinate through… See the full description on the dataset page:
https://huggingface.co/datasets/q-i-n-g/MECoBench.
