Edge AI-driven Multi-camera System for Adaptive Robot Speed Control in Safety-critical Environments
摘要
This paper proposes a safety system for speed control of collaborative robots using multiple cameras and edge AI. Collaborative robots share the workspace with human workers, which requires a high level of safety for the workers. ISO/TS 15066 is a safety regulation for collaborative robots, and the proposed system establishes an additional safety system beyond the built-in safety regulations of collaborative robots. Using the edge AI-based YOLOv5n-seg, the system detects workers and estimates the distance through the instance segmentation coordinates used in object detection. The distance accuracy through the RGB-D camera shows an error rate of 3.55% at 3 meters. The adjusted base joint’s speed is 0.09 rad/s in normal mode. This is about 1/6 of the original maximum speed of 0.6 rad/s. When applying the speed reduction rate of the safety system, a robot operating at maximum speed in normal mode complies with the speed and distance regulations of ISO/TS 15066. Additionally, the robot’s response time from object detection is 0.806 seconds, confirming that worker safety can be effectively ensured.