Real-time Collision Detection of Dual Rotating Arm Positioner for Multi-object Fiber-fed Spectrographs

Abstract Multi-object fiber spectroscopic survey is pivotal to astronomical research. Most spectroscopic telescopes are equipped with thousands of robotic fiber positioners designed to observe multiple celestial objects simultaneously. Despite this advancement, the risk of potential collisions between adjacent positioners, due to overlapping work zones, poses a significant challenge that could limit the telescope’s observing efficiency. In this study, we present a method based on deep learning to detect the collision of dual rotating arm positioner using the front-illuminated image from the Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST). We employ a deep learning detection model based on the YOLOv5 object detection algorithm to identify and locate the collision zones. Furthermore, the BiSeNet image segmentation algorithm is applied to determine the positioners within these collision zones, ultimately identifying the collided positioners. Experimental results reveal a precision and recall of 90.20% and 85.44% respectively for our method. To verify our results further, we conducted a correlation analysis on the spectral flux in LAMOST survey data via direct measurement. The collision types of the LAMOST positioners are also analyzed, which provides guidance for optimizing the anti-collision algorithm in the future.