MECoBench probes multimodal agent collaboration in embodied environments

arXiv·low signal

MECoBench is a multimodal embodied cooperation benchmark spanning diverse real-world tasks, two cooperation structures, and three collaboration modes, evaluating whether MLLMs can collaborate in visually grounded settings. Early findings show collaboration generally helps but is uneven across models and modes. Relevant to teams designing multi-agent MLLM systems that must coordinate rather than act solo.