An improved model based on cellular automata for on-line navigation

Abstract

Path planning and on-line navigation are some of the most-studied tasks in autonomous robotics. Since cellular automata (CA) are totally discrete models, they have also been recently considered for path planning. In order that the task be performed in a more decentralized manner, a local decision making approach is used in the CA-based models, where the next move is decided based on robot's sensors reading at each time step. Here, a simplified version of a previous navigation model to deal with single robot scenarios was adopted. This model is based on two sets of CA rules: those used to deviate from an obstacle identified within the robot's neighborhood; and those used to keep the robot navigating on a desirable axis whenever possible. However, by simulating this model some problems were noted in the robot's behavior. Two new strategies were implemented in the local decision making approach aimed at solving such problems. The continuous detection strategy is used to avoid collisions when the robot deviates from an obstacle, while a trajectory correction strategy aims to reduce the odometry errors arising from the robot's navigation. The refined model was simulated using the Webots platform and the results show an improvement in the robot performance, achieving a collision-free trajectory and a reduction from 55% in the errors with respect to its navigation axis.