Evaluating Multimodal Large Language Models with Daily Composite Tasks in Home Environments

TL;DR

This work introduces an embodied, daily-life benchmark for evaluating multimodal large language models (MLLMs) in a simulated home environment. Eight tasks spanning object understanding, spatial reasoning, and social activity are designed around early childhood milestones and tested with 17 MLLMs using a perception–reasoning–action loop and ReAct-style prompts in Unreal Engine 5. The study finds broad underperformance across all domains, highlighting a gap between current MLLMs and the general, embodied intelligence required for real-world tasks, and provides a framework for future development and standardized evaluation. The results emphasize the need for advances in multimodal perception, embodied reasoning, and social understanding to progress toward practical, embodied AI systems.

Abstract

A key feature differentiating artificial general intelligence (AGI) from traditional AI is that AGI can perform composite tasks that require a wide range of capabilities. Although embodied agents powered by multimodal large language models (MLLMs) offer rich perceptual and interactive capabilities, it remains largely unexplored whether they can solve composite tasks. In the current work, we designed a set of composite tasks inspired by common daily activities observed in early childhood development. Within a dynamic and simulated home environment, these tasks span three core domains: object understanding, spatial intelligence, and social activity. We evaluated 17 leading proprietary and open-source MLLMs on these tasks. The results consistently showed poor performance across all three domains, indicating a substantial gap between current capabilities and general intelligence requirements. Together, our tasks offer a preliminary framework for evaluating the general capabilities of embodied agents, marking an early but significant step toward the development of embodied MLLMs and their real-world deployment.

Figures (6)

• Figure 1: The illustration of eight task types.
• Figure 2: Overview of the eight task types with representative subtasks.
• Figure 3: System framework for evaluating MLLM-based embodied agents.
• Figure 4: MLLM agents' performance on eight tasks types.
• Figure 5: MLLM agents' performance on three domains.