A Comprehensive Study of Implementation Bugs in Multi-modal Agents

Multi-Modal Agents (M-agents), empowered by Large Language Models (LLMs), excel in various complex, open-world scenarios such as autonomous driving and robotics. However, their unique requirements to interact with dynamic and diverse multi-modal environments introduce novel implementation challenges...