Uma estratégia para navegação de robôs de serviço semiautônomos usando informação local e planejadores probabilísticos

Abstract

Service robots are designed to perform useful tasks for humans, such as cleaning environments, delivering an order or assisting human mobility, especially, assisting elderly or people with disabilities. Among several types of service robots, this work is focused in the robots classified as semi-autonomous service robots. These robots are partially controlled by a person, being the robot responsible for specific tasks, especially those related to mission safety. This master thesis proposes a strategy for the navigation of semi-autonomous service robots based on local information obtained by the sensors installed on the robot and an probabilistic planner, which calculates paths to be followed locally by the robot, thus ensuring obstacle avoidance. The robots mission is defined by a user through a sequence of simple commands, such as go aheadand turn right. These commands are encoded by artificial vector fields, which are used by the functional cost to be optimized by the planner. This navigation strategy is interesting because semi-autonomous service robots, unlike others, may not have a map of the environment or a global localization system, since they may be navigating in unknown environments under close supervision of a human. Purely deliberative architectures use path planners to avoid obstacles and usually require the localization of the robot on a map, being difficult to be applied to this type of robot. On the other hand, purely reactive architectures may take the robot to undesired conditions or may generate non-smooth paths. For the experimental evaluation of the proposed navigation strategy, we considered a typical environment of office buildings, hospitals, and schools that is consisted of corridors and intersections. In this environment, the strategy was used to control a simulated robot and two actual semi-autonomous service robots. The results show that the robots were able to perform the user-defined mission efficiently, avoiding static obstacles and, for some well-defined situations, also dynamic obstacles, including people moving in the environment during the mission. The comparison with classical methods based on SLAM shows that, besides simpler and more computationally efficient, the proposed methodology presents better results in dynamic environments. When comparing with a reactive strategy that uses potential fields, it is noticed that this strategy does not always provide smooth paths, and can lead to local minimum conditions, unlike the result obtained with the our methodology.