• collision
• path planning
• risk map
• uncertainty

The presence of uncertainty in the robot configuration and that of the other static or moving entities in the surrounding (e.g. human beens or obstacles) can result in more frequent and severe collisions for an autonomous mobile robot operating in an indoor, dynamic, and unknown environment. To address this problem, this thesis develops a probabilistic strategy for assessing and managing collision risk due to uncertain sensing. The uncertainties considered arise from two different sources: laser range finder measurement noise and pose estimation errors. The integration of these uncertainties, having Gaussian probability density functions, into two bidimensional grid maps results in an accurate representation of the probability with which static objects and dynamic entities, respectively, occupy the workspace. Since a collision can be defined as the intersection of two or more occupancy events, from the knowledge of the actual occupancy grid maps two bidimensional risk grid maps are constructed containing the probability with which, at the considered time instant, a collision between the robot and the static and dynamic entities takes place. Two scaling factors having exponential trends are introduced. The first takes into account the increase in uncertainty that occurs when previously explored portions of the environment are no longer visited for a long time due to occlusions or limited sensor range. The second takes into account lower reliability of a predicted collision when the time instant in which it is expected to occur is far from the current one. A collision risk-RRT* based on the generated grid maps is implemented for minimum-risk planning in a changing and uncertain environment. The cost function represents the average collision risk on the path. The collision risk-RRT* plans offline a path that the robot executes by serving a Proportional-Integral-Derivative feedback controller. Reaching each waypoint involves enabling a checker that by estimating the probability with which, in a given prediction horizon, a collision takes place, determines whether or not an online replanning is necessary. The collision risk-aware mapping and path planning algorithms under uncertainties are developed in C++ and tested on ROS using the Robotnik-designed robotic platform SUMMIT-XL STEEL.