FAM-HRI: Foundation-Model Assisted Multi-Modal Human-Robot Interaction Combining Gaze and Speech

Effective Human-Robot Interaction (HRI) is crucial for enhancing accessibility and usability in real-world robotics applications. However, existing solutions often rely on gesture-only or language-only commands, making interaction inefficient and ambiguous, particularly for users with physical impairments. In this paper, we introduce FAM-HRI, an efficient multimodal framework for HRI that integrates language and gaze inputs via foundation models. By leveraging lightweight Meta ARIA glasses, our system captures real-time multimodal signals and utilizes large language models (LLMs) to fuse user intention with scene context, enabling intuitive and precise robot manipulation. Our method accurately determines the gaze fixation time interval, reducing noise caused by the gaze dynamic nature. Experimental evaluations demonstrate that FAM-HRI achieves a high success rate in task execution while maintaining a low interaction time, providing a practical solution for individuals with limited physical mobility or motor impairments. To support the community, we have released our system design, algorithms, and solutions at https://github.com/laiyuzhi/FAM-HRI Note to Practitioners—In the field of assistive and service robotics, enabling users with physical disabilities to communicate intentions naturally remains a significant challenge. Existing uni-modal approaches, such as voice-only or gaze-only interaction, often struggle with ambiguity and environmental interference, limiting their reliability in real-world scenarios. Current multimodal systems typically process gaze and speech separately, making it difficult to align visual attention with oral commands accurately. Additionally, many systems rely on bulky hardware or require users to maintain prolonged focus, reducing practicality and comfort. To address these issues, this paper proposes a foundation-model-assisted multimodal human-robot interaction (FAM-HRI) framework that combines lightweight AR glasses with LLMs to fuse gaze and speech inputs in real time. The system automatically filters noisy gaze signals, aligns them with spoken commands, and generates precise robot actions. Experiments demonstrate that FAM-HRI achieved task success rates exceeding 94% across multiple scenarios and outperformed all baseline methods in interaction efficiency, with the shortest interaction times in comparative evaluations. This method is well-suited for applications such as household assistance, and collaborative tasks where hands-free interaction is desirable. In the future, we plan to optimize computing performance and explore strategies to resolve conflicts between gaze and speech inputs, further enhancing usability and adaptability in diverse environments.