An Approach for Opening Doors with a Mobile Robot Using Machine Learning Methods

One of the tasks robotics is to develop a robot’s ability perform specific actions for as long possible without human assistance. such step open different types doors. This task essential any operation that involves moving robot from one room another. paper proposes versatile and computationally efficient algorithm an autonomous mobile opening doors, using machine learning methods. The latter include YOLOv5 object detection model, RANSAC iterative method estimating mathematical model parameters, DBSCAN clustering algorithm. Alternative methods are also compared. proposed was explored tested in simulation on real manufactured by SOMATIC version Dalek. percentage successful doors opened out total number attempts used accuracy metric. reached 95% 100 attempts. result testing door-handle simulated data error 1.98 mm 10,000 samples. That is, average distance door handle found detector mm. has shown high be applied time