Robot Navigation in a Dynamically Changing Environment

This study considers the operation of an onboard navigation system, which determines the robot position in the environment. Navigation system (NS) errors arising (accumulated) in the process of operation from errors of accelerometers and angular velocity sensors shall be compensated by certain correction allowing to refine the robot coordinates. Various approaches are used to correct NS, in particular, visual navigation methods implemented in onboard computer vision systems. Most of the approaches use neural networks based on a computer vision system, which functions under the conditions of direct visibility of reference points. The disadvantages of methods with special markers include the need to install them in different locations of the room (on the walls, floor), which requires additional time and resources to organize the required information field. Furthermore, such methods are ineffective in a dynamically changing environment, in particular, in crowded rooms, where moving people or transport robots can obstruct reference points and disrupt NS operation. The approach proposed by the authors considers a scheme for integrating navigation data coming from the inertial measurement unit (IMU), encoder and computer vision system, and is based on the extended Kalman filter (EKF). Coordinate correction is carried out by a computer vision system, which matches the current images of the ceiling sections obtained by the robot to a previously prepared image of the entire ceiling. Matching is implemented by an extreme correlation algorithm for image matching. The proposed system allows solving robot's navigation tasks in a changing dynamic environment, in particular, with people moving around the robot. Experimental studies of the correlation algorithm for estimating coordinates have shown the centimeter accuracy of the visual coordinate estimation algorithm. Modeling the coordinate estimation process using the Kalman filter confirms the system's effectiveness, including cases of failure of one or more navigation devices.