M.Sc Thesis


M.Sc StudentAvraham Turgeman
SubjectSensor Data Fusion of a Redundant Dual Platform Robot for
Terrain Mapping
DepartmentDepartment of Robotics and Autonomous System
Supervisors Assistant Professor Degani Amir
Dr. Shoval Shraga
Full Thesis textFull thesis text - English Version


Abstract

This research presents a new technique of terrain mapping along a defined course using a redundant UGV (Unmanned Ground Vehicle) robot. The robot is designed to work in unknown and challenging environments such as confined spaces, tunnels, pipes and inside buildings, without the use of external information, which often is unavailable due to lack of communication with external devices such as GPS and beacons. The robot design consists of two platforms connected with a rigid mechanism. This connecting mechanism is a beam with two connecting bars. All joint positions in the mechanism are constantly measured by potentiometers to retrieve joint angles. This mechanism allows relative motion measurements of each platform in relation to the other platform. In addition, each platform includes on-board sensors, i.e., inclinometers, accelerometers and gyroscopes, providing sufficient data for accurately and independently determination of the platform orientation. All of the above turns the robot into a six-degrees-of-freedom (DOF) dynamic system with redundancy, which allows control and stabilization of the platform, as well estimation of the terrain features. According to the dynamic patterns of the redundant information, a mathematical algorithm produces a map by performing data fusion.

Several methods had been tested. Due to sensors’ noise, sensitivity and bias, and the non-linear relations between the state vector parameters, the Extended-Kalman-Filter (EKF) method was chosen. For resolving the multi-measurements problem, an upgrade of the EKF is advised. This optimal estimator, named Centralized Kalman Filter (CKF), provides a “trust-parameter” for each data measured in each step by taking into consideration the previous results for enhanced estimation of the current step. Finally, the algorithm generates a 2-D map. The proposed algorithm was initially tested in simulation. Many scenarios were tested using the algorithm - from ideal world (without any measurement or process errors) to several errors in odometry model and inertial measurements. During the tests, some algorithm improvements were added for better results. The robot was tested in varies terrains. From indoor, flat and well controlled environment, to rough, challenging gravel oval terrain. The research compared between different map results. Single robot, leading platform and CKF algorithm produced maps according to the terrain, and several parameters were tested. Simulation validation was also conducted to assure the experimental results. This will enable verification of robot mapping results in additional terrains. The research focused on two-dimensional map only while driving along a single path.