Remote Human-Robot Interaction in Industrial Inspection System Based on Vision-Language Models

To address the challenges of complex operations, high personnel training costs, and limited system flexibility and scalability in industrial inspection system, this paper proposes a remote human-robot interaction scheme in industrial inspection system based on a vision-language model (VLM). This scheme allows operators to issue inspection task commands remotely using natural language, such as text or voice. The commands can range from detailed operational requirements to simple action refinements, such as "Move to the first instrument on the left side of the image." or "Move slightly to the left." Upon receiving a command, the VLM comprehensively analyzes the instruction in conjunction with image data collected by the robot's visual sensors and environmental perception data provided by LiDAR. This process effectively eliminates command ambiguity and generates accurate control instructions for the robot to ensure precise execution of the inspection tasks. Experimental results demonstrate that this scheme maintains a high success rate and stability.