Hybrid genetic algorithm and particle filter optimization model for simultaneous localization and mapping problems
Determining position of a robot and knowing position of the required objects on the map in unknown environments such as underwater, other planets and the remaining areas of natural disasters has led to the development of efficient algorithms for Simultaneous Localization and Mapping (SLAM). The curr...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Language: | English |
Published: |
2016
|
Subjects: | |
Online Access: | http://eprints.utm.my/id/eprint/77753/1/MohsenMahramiPFC2016.pdf http://eprints.utm.my/id/eprint/77753/ http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:97493 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Determining position of a robot and knowing position of the required objects on the map in unknown environments such as underwater, other planets and the remaining areas of natural disasters has led to the development of efficient algorithms for Simultaneous Localization and Mapping (SLAM). The current solutions for solving the SLAM have some drawbacks. For example, the solutions based on Extended Kalman Filter (EKF) are faced with limitation in non-linear models and non-Gaussian errors which are causes for decrease of accuracy. The solutions based on particle filter are also suffering from high memory complexity and time complexity. One of the major approaches to solve the SLAM problem is the approach based on Evolutionary Algorithm (EA). The main advantage of the EA is that it can be used in search space which is too large to be used with high convergence while its disadvantage is high time and computational complexity. This thesis proposes two optimization models in solving SLAM problem namely Hybrid Optimization Model (HOM) and Lined-Based Genetic Algorithm Optimization Model (LBGAOM). These models do not have the limitations of EKF, memory complexity of particle filter, and disadvantages of EA in search space. When the results of HOM compared with original EA, it showed an increase of accuracy based on presented fitness function. The best fitness in original EA was 16.36 but in HOM has reached to 16.68. Both models applied a proposed new representation model. The representation model is designed and used to represent the robot and its environment and is based on occupancy grid and genetic algorithm. There are two types of representation models proposed in this thesis namely Layer 1 and Layer 2. For each layer, related fitness function is created to evaluate the accuracy of map in the model that was tested with some different parameters. The proposed HOM is designed based on genetic algorithm and particle filter by creating a new mutation model inspired by particle filter. The search space is reduced and only suitable space will be explored based on proposed functions. The proposed LBGAOM is a new optimization model based on extraction line from laser sensor data to increase the speed. In this model, search space in the map is a set of lines instead of pixel by pixel and it makes searching time faster. The evaluation of the proposed representation model shows that Layer 2 has better fitness value than Layer 1. The HOM has better performance compared to original GA Layer 1. The LBGAOM has decreased the search space compared to pixel based model. In conclusion, the proposed optimization models have good performance in solving the SLAM problem in terms of speed and accuracy. |
---|