SignVLA: Real-Time Sign Language-Guided Robotic Manipulation via Attention LSTM and Vision-Language-Action Models

SignVLA is a newly introduced framework that enables real-time robotic manipulation guided by sign language, utilizing an attention-enhanced Long Short-Term Memory (LSTM) network for gesture recognition. The system processes video streams to extract hand landmark features, translating sign gestures into semantic instructions for Vision-Language-Action (VLA) models, thus enhancing accessibility for users with speech impairments. Experimental results indicate that SignVLA achieves stable real-time sign recognition and effective manipulation task execution, highlighting its potential as an accessibility layer in multimodal robotic systems.