Motion Control Of Floor Curling Robot Based On Semantic Image Segmentation Algorithm

With the development of artificial intelligence,more and more applications of artificial intelligence appear in our life.Curling robot is the product of the combination of artificial intelligence technology and robot technology,with the goal of completing the curling competition task through the cooperation of decision-making,throwing,rubbing ice and so on.In this process,the robot involves mechanical structure design,visual perception processing,decision-making and control,which is also the most complex role.Under this background,this paper studies the motion control of the floor curling robot based on semantic segmentation algorithm,and simulates the motion of the ice wiping robot on the curling field.The hardware structure of the floor curling robot mainly includes three parts:vision module,chassis module and control center.The vision module is used to obtain the external environment information from the perspective of the floor curling robot.The chassis module realizes the specified movement by controlling the Mecanum wheel.The control center carries out semantic segmentation of the visual information,and gives the chassis control instructions to complete three motions: waiting,following and returning.Firstly,this paper studies the semantic segmentation algorithm.The motion control of floor curling robot requires real-time,so the complexity and reasoning time of semantic segmentation algorithm need to be considered comprehensively.This paper compares the classic semantic segmentation networks such as FCN,Seg Net,UNet and ENet,and improves a real-time semantic segmentation network DFANet by compressing the structure of dfanet and introducing multithreading acceleration,the algorithm can be used in real time without losing a small amount of segmentation accuracy.It ensures that the algorithm can be used in real time with a small loss of segmentation accuracy.Secondly,the motion control of Mecanum wheel chassis is studied.Firstly,the kinematics principle of Mecanum wheel is analyzed.Aiming at the problems of dead time and unbalanced movement of four-wheel speed command in Mecanum wheel motor,the mapping relationship between speed command and PWM signal is studied.Aiming at the problem of insufficient parameters in the mapping relationship and unable to directly measure Mecanum wheel speed,a parameter measurement method based on observing chassis motion state is proposed,The robot can move accurately in all directions.Finally,the following control problem of floor curling robot based on vision sensor is studied.In this paper,the closed-loop problem of robot position is transformed into the closed-loop problem of curling pixel coordinates in the image.At the same time,limited to the speed of semantic segmentation,PID method with non-uniform time interval is used to control the robot.The semantic segmentation algorithm and PID follow-up control algorithm are transplanted to the floor curling robot platform to realize the three motion modes of the floor curling robot: waiting,following and returning.