DefinePK

Abstract

The design of robotic platforms for service robotics applications requires a good solution to the problem of autonomous mobility. One of the basic tasks of these robots is to navigate through structured environments with large displacements while traveling a predefined path. A good solution to this problem is the visual identification of the path and characteristics of the environment (free space and obstacles) using digital cameras and processing of its information to infer the possible movements of the robot. In this paper, we propose a navigation scheme based on visual feedback formed by two control loops. The first control loop corresponds to a visual memory strategy based on a convolutional neural network. The training of the control system is performed using images with characteristic elements of the navigation environment (different obstacles and different cases of free paths, with and without a navigation path). The system is propagated over the robot using the current images of two cameras in stereoscopic configuration. The second control loop defines specific distances to the obstacles using stereoscopic vision. This second control loop aims to quickly identify points of the obstacles in front of the robot from the images using a bacterial interaction model. These points are mapped in the planes of projection of the environment to establish the distance to the obstacle. Here, we introduce a complete framework that efficiently performs stereoscopic vision-based analysis in real-time, by significantly reducing the computational cost. Our strategy does not require modifications to the environment, nor special camera schemes in the robot, so it can be used in many robot topologies and many different environments without major adjustments. Simulations and real-world experiments show the validity and performance of the proposal.