Title:

Leveraging A Large Language Model for Movement Intention Recognition

Current movement intention recognition systems rely on sensors such as electromyography sensors (EMG) and inertial measurement units (IMU) which rely on a change in movement to predict an intended movement. With recent advancements in large language models (LLMs) capable of analyzing visual data, there is potential to explore alternative approaches. This study investigates the feasibility of using LLMs for movement intention recognition by extracting frames from point-of-view videos of various movements, stitching these frames into a single image, and inputting the image into an LLM for interpretation. Experimental results show that the LLM achieved recognition accuracy between 70% and 85%, with average response times ranging from 5.5 to 7 seconds. Performance was notably influenced by the number of frames selected and the duration of the source video. While the approach demonstrates promise, the current speed and accuracy suggest that LLM-based video input systems may not yet be practical.

Name

Comment*

Comments are viewable only by submitter

Captcha*