Few-Shot Visual Grounding for Natural Human-Robot Interaction
This work addresses natural human-robot interaction for service robots in dynamic environments, presenting an incremental improvement by combining zero-shot learning with visual grounding.
The paper tackles the problem of enabling robots to understand human verbal commands to segment target objects in crowded scenes, proposing a single-stage zero-shot visual grounding model that achieves good accuracy and speed on real RGB-D data.
Natural Human-Robot Interaction (HRI) is one of the key components for service robots to be able to work in human-centric environments. In such dynamic environments, the robot needs to understand the intention of the user to accomplish a task successfully. Towards addressing this point, we propose a software architecture that segments a target object from a crowded scene, indicated verbally by a human user. At the core of our system, we employ a multi-modal deep neural network for visual grounding. Unlike most grounding methods that tackle the challenge using pre-trained object detectors via a two-stepped process, we develop a single stage zero-shot model that is able to provide predictions in unseen data. We evaluate the performance of the proposed model on real RGB-D data collected from public scene datasets. Experimental results showed that the proposed model performs well in terms of accuracy and speed, while showcasing robustness to variation in the natural language input.