Active slamutility functions and applications

Resumen

The use of robots in applications that are useful for human beings can have a great impact on our quality of life and in our economy. Using robots in tasks that are hazardous (e.g. humanitarian demining or nuclear plants inspection) or repetitive for humans (e.g. area surveillance or inventorying) can enhance our life quality. Similarly introducing robots in factories can boost the efficiency in many industrial tasks. A robot that performs useful tasks for humans is known as a service robot and, according to its standardized definition, a robot needs to show some degree of autonomy while performing its task, i.e. the robot should not require complete human intervention to achieve its task. Granting autonomy to a mobile robot requires, among other things, solving tree basic tasks in robotics, namely, localization, mapping and motion control. Localization, refers to the robot answering the question "Where am I?". Mapping, refers to the robot answering the question "How does the world look like?". Finally, motion control, refers to the robot answering the question "How do I go from point A to point B?". The solution of these three basic tasks are interlaced, knowing where the robot is located, i.e. localization, requires a map of the environment. Constructing a map of the environment, i.e. mapping, requires to know the position where the sensor data of the environment was acquired. Moving the robot to one pose to another, requires both localization and mapping. Moreover, in order to perform a task, a service robot needs an operative model of the environment. If the task has to be done without human intervention, i.e. full autonomy, the robot should at least be endowed with the capability of solving the three aforementioned problems: localization, mapping and motion control. Solving these three problems jointly in order to obtain the most accurate representation of the environment in a finite time is known as active Simultaneous Localization and Mapping (active SLAM), and it is the main theme of this thesis. Active SLAM refers to a solution of the robotic exploration problem in which the main concern of the robot performing the exploration task is to obtain the most accurate representation of the environment, i.e. uncertain as well as unknown parts of the environment are of the utmost interest. In this setting, the robot must make a trade-off between exploring new area to complete the task and exploiting the existing information to maintain good localization, i.e. solving the so-called exploration-exploitation dilemma. Selecting actions that decrease the map uncertainty while not significantly increasing the robot's localization uncertainty is challenging. Active SLAM algorithms select the next robot¿s action based on its utility to solve the exploration-exploitation dilemma. In this thesis we focus on utility functions for active SLAM algorithms. We analyze state-of-the-art utility functions in the context of active SLAM, and in particular we delve into their components related to quantifying the uncertainty of the robot and the map. The utility functions are of paramount importance in an active SLAM algorithm as they decide the next action of the robot. Therefore, a faulty utility function can lead the robot to take actions that most likely will jeopardize the achievement of the robot¿s high level goals. The main contributions of this thesis can be briefly summarized as follows: - We present an analysis of the application of optimality criteria stemming from the Theory of Optimal Experimental Design in order to quantify the uncertainty in both the robot and the map for active SLAM. Moreover, we provide a correction to the formulae used in the robotic community to quantify uncertainty using the aforementioned optimality criteria in an active SLAM scenario. - We present a novel information-theoretic utility function for active SLAM that uses both Shannon's and Rényi's definitions of entropy to jointly consider the uncertainty of the robot and the map. This allow us to fuse both uncertainties without the use of an ad hoc solution.