Close

Presentation

41. High-Fidelity Worker Motion Simulation With Generative AI
DescriptionThis research proposes a dual-phase AI-based methodology to enhance worker motion simulation within industrial settings by addressing the persistent challenges of accuracy and fidelity. Initially, it utilizes Large Language Models (LLMs) to distill instructions through Prompt distillation, incorporating texts and frequent keywords within a limited action dataset used for training a VQ-VAE-based text-to-motion model and generating alternative motion descriptions. In the second phase, temporal similarity metrics are employed to validate the generated motion against real human motion specific to the task. This comparison provides an estimate of the discrepancy between these two sequences, which can be further minimized by maximizing their similarity. Subsequently, precise motion generation is achieved from these instructions using a VQ-VAE-based text-to-motion generator model. The proposed methodology is valuable for its adaptability to datasets lacking extensive variations in actions, providing a solution for enhancing worker training and efficiency within manufacturing environments.
Event Type
Poster
TimeThursday, September 12th5:30pm - 6:30pm MST
LocationMcArthur Ballroom
Tracks
Aging
Augmented Cognition
Children's Issues
Communications
Cybersecurity
Education
Environmental Design
General Sessions
Human AI Robot Teaming (AI)
Macroergonomics
Occupational Ergonomics
Student Forum
Surface Transportation
Sustainability
System Development