An unified framework for active SLAM and online optimal motion planning

In this paper we propose an unified approach for active SLAM (Simultaneous Localization And Mapping) and optimal motion of nonholonomic vehicles. Both the environment and the vehicle model are unknown in advance, so the path planner uses reinforcement learning to acquire the vehicle model, which is estimated by a reduced set of transitions. At the same time, the vehicle explores the environment creating a consistent map through optimal path motion. The mapping is represented by a set of ordered and weighted particles, named objects, that provides some important advantages with respect to conventional methods. In order to guide the navigation and to build a map of the environment the planner employs a three-dimensional controller based on the concept of virtual wall following. Both simulation and experimental results are reported to show the satisfactory performance of the method.