This research advances beyond conventional IMU-based activity recognition by fine-tuning large language models to understand subtle human movements like air-written letters.

• Current LLMs fail at fine-grained movement recognition tasks (near-random accuracy)
• Researchers developed a specialized fine-tuning approach for detailed movement patterns
• Opens new possibilities for security applications including enhanced biometric authentication and more precise motion-based identification
• Creates foundation for advanced human-computer interaction through subtle gesture recognition

Security implications include more sophisticated movement-based authentication systems and improved detection of unusual behavior patterns in security monitoring applications.

Exploring the Capabilities of LLMs for IMU-based Fine-grained Human Activity Understanding