Abstract of Thesis presented at COPPE/UFRJ as a partial fulfillment of the requirements for the degree of Master of Science (M.Sc.)

Localization of an Autonomous Rail-Guided Robot Based on Particle Filter

Guilherme Pires Sales de Carvalho

March/2016

Advisor:

Ramon Romankevicius Costa

Department:

Eletrical Engineering

      Recently, it has b een noted a growing interest on rob otic systems in pro duction facilities of the oil and gas industry, esp ecially on offshore platforms . One problem in particular of those facilities is the maintenance of pro cess plants, which currently requires flying human op erators to distant fields to p erform insp ection and maintenance tasks on site, b eing sub ject to the risks inherent to the characteristic harsh environment and accounting costs related to complex logistics.
      DORIS is a rob ot b eing develop ed by COPPE in collab oration with Petrobras and Statoil as a solution to this problem, with the purp ose of autonomously carrying out insp ection tasks in offs hore platforms pro cess plants. The rob ot moves through a rail installed in the regions of interest of the plant and it is equipp ed with various sensors capable of providing real time information ab out the insp ected environment to a remote op erator.
      This work presents the implementation of a lo calization algorithm for the autonomous navigation system of DORIS using a probabilistic approach. A particle filter, which integrates motion and p erception information of the rob ot, is used to estimate the rob ot lo calization on the rail. A novel technique, based on the recent history of events observed by the rob ot, is prop osed to augment the standard Monte Carlo lo calization algorithm to improve the robustness and convergence rate of the estimation, and reduce its computational complexity.
      Simulations using field tests data of DORIS show that the prop osed algorithm estimates the rob ot p osition on the rail with an error smaller than 25cm, proving to b e sup erior than o dometry or a standard particle filter. In addition, the algorithm is able to solve imp ortant lo calization problems for autonomous rob ots, which are the initial condition, the global lo calization, and the kidnapp ed rob ot problems.