This research introduces a comprehensive benchmark for evaluating how well multi-modal AI models understand the world from a first-person perspective—critical for advancing embodied AI and robotics.

Key Innovations:

• Addresses critical gaps in existing embodied cognition evaluation frameworks
• Tests AI's ability to handle robotic self-cognition and dynamic scene perception
• Evaluates and mitigates hallucination in vision-language models processing egocentric data
• Creates a systematic assessment for embodied cognitive abilities in large vision-language models

For engineering applications, this benchmark enables more reliable development of robots that can accurately perceive and interact with their environment from a first-person perspective, potentially improving safety and functionality in human-robot collaboration scenarios.

ECBench: Can Multi-modal Foundation Models Understand the Egocentric World? A Holistic Embodied Cognition Benchmark