Visual control system design of wheeled inverted pendulum robot based on Beaglebone Black

Abstract

The wheeled inverted pendulum robot has broad prospects of applications in real life. It can use two coaxial wheels to achieve the body self-balancing, forward moving and turning. But the general wheeled inverted pendulum robot seldom has vision function to perceive enviromental change. In order to realize the robust visual control, a wheeled inverted-pendulum vision robot with attitude sensors, photoelectric encoders, ultrasonic sensors and so on is designed based on Beaglebone Black board. The moving object is separated in the space domain by obtaining the image sequence which is sent by a robot-mounted camera, and the modeling, identification and tracking of target sequence are implemented in the time domain. The balance PD, speed PI and steering PD controllers are designed to realize the dynamic balance, forward and steering function of the robot. To satisfy the functional requirements of the visual tracking system, an improved tracking-learning-detection algorithm based on kernelized correlation filtering is used, and a tracking anomaly based on spatial context is detected to determine the tracking state and reduce the error rate. Experimental results show that the robot reaches the requirement of design and achieves better visual control effectiveness.