Integrating Speech and Gesture for Generating Reliable Robotic Task Configuration

摘要

This paper presents a system that combines speech and pointing gestures along with four distinct hand gestures to precisely identify both the object of interest and parameters for robotic tasks. We utilized skeleton landmarks to detect pointing gestures and determine their direction, while a pre-trained model, trained on 21 hand landmarks from 2D images, was employed to interpret hand gestures. Furthermore, a dedicated model was trained to extract task information from verbal instructions. The framework integrates task parameters derived from verbal instructions with inferred gestures to detect and identify objects of interest (OOI) in the scene, essential for creating accurate final task configurations.