SENSOR BASED MOBILE ROBOT NAVIGATION

Abstract

Sensor based motion planning and world modelling are two major tasks for mobile robot navigation. A sensor based motion planning strategy allows mobile robots to navigate autonomously in an area without prior knowledge of the obstacle locations. World modelling is a process of creating a map of the environment in which the mobile robot navigates. In this thesis, a new strategy for sensor based motion planning of a mobile robot in an area without prior knowledge is developed. The algorithm converts the current ultrasonic sensor information into virtual forces. The virtual forces are used to guide the robot and navigate it to its destination while avoiding the potential risk of collision with obstacles. The computational expense of implementing this algorithm is low. It is also robust and is able to adopt to any kind of environment. Real-time management of sensor information and world modelling is another important research topic related to sensor based mobile robot navigation. World modelling is one of the objectives of mobile robot navigation. In this thesis, a new method is presented which uses incremental Least Squares Method (LSM) to approximate obstacle boundaries as line segments using real-time sensor information. These line segments form the boundaries of the obstacles which compose the world map and are geometrical representations of the boundaries of the obstacles. This map is called Partially Finished Geometrical Map (PFGM). In order to obtain a 2D world model which represents all objects in an area, the complete exploration of the whole unknown region is necessary, i.e., the mobile robot must explore the whole area. A new algorithm is developed and introduced in this thesis for the complete exploration of the region. The map building process and the mobile robot navigation process are complementary and run simultaneously. The Partially Finished Geometrical Map (PFGM) is built incrementally and object with incomplete map are selected from the PFGM. The exploration process continues until all objects inside a certain region have been detected and built into the map. A 2D world model which represents all objects in the region is obtained through this method. The simulation and experimental results are also presented in this thesis for all the above methods using the mobile robot test bed - Nomad 200.