The probability estimation problem of a collision between path tracking for an autonomous mobile robot with an obstacle is considered. We reviewed and analyzed methods for solving this problem. We show that reviewed methods use periodically updated grid maps (occupancy grids). The new method of probability estimation of the collision between the mobile robot with an obstacle is presented. This method based on the use of probabilistic grid map. Each cell of this map stores the estimated probability that the obstacle is located within. In addition, this map stores the conditional probability of occupying of the map cells by a robot, taking into account the possible lateral and angular deviation from the planned trajectory. This deviation caused by error connected with dynamic characteristics of the tracking system. To build the probabilistic occupancy grid, the dynamically updated multilayer grid map was used. Each layer of this map, except for the resulting output, has been filled with the data obtained from classifiers which process information incoming from sensory of the robot. This layer is the result of Bayesian inference from the layers laying below. The motion control system provides construction of the multilayered grid maps, probabilistic occupancy grids, coordinate estimations, path planning, motion tracking and the probability estimation for collision with obstacles. The method such estimation is implemented as an embedded module compatible with ROS (Robot Operating System). The description of experiments with the mobile robot in-nature (on the field) is given in the case when a motile obstacle appears intercepting the planned path. The estimated changes of probability for a collision between the mobile robot with obstacle are presented, interpretation of the obtained results is also given. Here we demonstrated the necessity of collision probability estimation for assessment of the risk as the main safety indicator of the given motion control system. Results of this work are considered and evaluated as a solution to the problem of ensuring the safety of motion tracking for autonomous mobile robots.
A Deep Learning Approach for the Mobile-Robot Motion Control System
