Improving path planning of autonomous vacuum cleaners using obstacle classification

Abstract: To effectively plan their movement, robotic vacuum cleaners require information about their surrounding environment. This thesis presents a mapping algorithm which uses collision input from a moving robot to create a grid map of the environment. The algorithm also classifies each obstacle cell in the grid map as either mobile or immobile. Methods for creating maps of unknown environments already exist. However, they often rely on information from expensive sensory equipment, and the maps do not include information about the mobility of the objects in the map. Hence, the aim of this thesis was to investigate if the map could be created using only limited information from more accessible sensors. A testing environment was created using the Unity 3D game engine, in which the robot was represented by a circular object. The robot had access to perfect information about its current position in relation to its starting position, the direction in which it was heading and any incoming collisions. Three test scenes were created in the simulation environment, representing different common operating spaces for a robotic vacuum cleaner. The scenes contained different kinds of mobile and immobile obstacles that the robot collided with. A series of tests were then conducted in the test scenes. The tests examined the performance of the created algorithm. The results indicated that the algorithm can create a grid map representation of an unknown environment and classify objects within it with an average correctness of around 80%. However, it is hard to say whether the algorithm would be effective in a real situation, due to the inconsistent results and unrealistic assumptions.