Uncertainty-based localization in a topological robot navigation system

Abstract

This thesis is distributed under license "Creative Commons Attribution -Non Commercial -Non Derivatives".To the past, my childhood, to the present, my family, and to the future.

Topological Robot Localization in a Large-Scale Water Pipe Network

“…Between junctions there is a chance that the robot is turned around and returns to the previous junction. When arriving at a junction there is a chance that the robot does not detect it, and continues moving in a random unknown direction, without updating the state estimate [11]. This model is illustrated as a discrete probability distribution shown in Fig.…”

“…As the problem is considering only localization in a known environment, rather than mapping, the possible transition and measurement models can be precomputed for a given network [11].…”

“…Recent work on topological localization adds the challenges of erroneous repeated observations of the environment at a topological map node, inclusion of information assigned to nearby nodes, and failing to make an observation at a node [11], the last of which is especially applicable where a robot has limited sensing ability as is the case in a pipe environment. Use of geometric information on the robot's orientation has been applied using prior knowledge of the orientation between two topological map nodes [12].…”

Adding Uncertainty to an Object Detection System for Mobile Robots

“…In order to show the applicability and to evaluate the efficiency of our proposed method, we integrate it with the topological navigation system presented in [51]. The test has been carried out in the apartment 1 of the Bosch dataset using the Gazebo simulator.…”

“…Third, we conduct a qualitative and quantitative evaluation on the Bosch Semantic Interpretation Challenge dataset [16] obtaining better performance concerning a baseline method. Besides, we integrate the global search method with a topological navigation system [51] for performing the optimized search in order to show the benefits and feasibility of implementing our approach in mobile robots. Regarding the local strategy, the main contributions are: first, a probabilistic analysis and selection of best viewpoints to find objects in partially known environments efficiently.…”