Design and development of cooperative simultaneous localization and mapping integrated with neural network for mobile robot
Autonomous cooperating mobile robots is one of technological advancement in enabling autonomy in search and rescue task (SaR). While, simultaneous localization and mapping (SLAM) algorithm is one of a key element to enable autonomous navigation. Many researchers chose to adopt high-end sensors to so...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2024
|
| Online Access: | http://eprints.utem.edu.my/id/eprint/28261/1/Design%20and%20development%20of%20cooperative%20simultaneous%20localization%20and%20mapping%20integrated%20with%20neural%20network%20for%20mobile%20robot.pdf http://eprints.utem.edu.my/id/eprint/28261/2/Design%20and%20development%20of%20cooperative%20simultaneous%20localization%20and%20mapping%20integrated%20with%20neural%20network%20for%20mobile%20robot.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Autonomous cooperating mobile robots is one of technological advancement in enabling autonomy in search and rescue task (SaR). While, simultaneous localization and mapping (SLAM) algorithm is one of a key element to enable autonomous navigation. Many researchers chose to adopt high-end sensors to solve Cooperative SLAM or CSLAM. But this will cause staggering cost for multiple robot system. Thus, the alternative is to implement low-cost sensors with limited sensing to perform CSLAM. However, this approach introduces challenges such as inaccurate robots’ sensors measurements and low accuracy cooperative mapping were reported. In this research, an Artificial Neural Network (ANN) is proposed to improve the accuracy of the CSLAM algorithm with lowcost sensors and is evaluated using real robots. Here, the selected methodologies are divided into three important stages to support three objectives defined to solve the stated problem statement. Firstly, the ANN configurations is established to reduce the nonlinearity error of the low-cost sensor measurements for building high accuracy environmental map. By training the ANN using sensor measurements, it learns to model the data and reduce the error or uncertainties present in the measurements obtained from the low-cost sensors. Secondly, a framework of CSLAM algorithm integrated with ANN using Rao-Blackwellized particle filter (RBPF) algorithm for single SLAM robot, and the map merging using random sample consensus (RANSAC) algorithm, is designed and developed. Lastly, the performance of the CSLAM algorithm with ANN is evaluated and validated using measurements from real robot platforms, and compared to that without ANN. From the real-world experiment, CSLAM with ANN has increased the performance of resulting maps by 61.09% compared to without ANN. It shows that, CSLAM integrated with ANN have improved the performance of CSLAM significantly. Moreover, CSLAM integrated with ANN have achieved 3 closed loop condition out of 10 trials for 600 particles compared to without ANN that does not achieve closed loop map out of 10 trials even though, the number of particles is increased. From the results, it can be concluded that the development of CSLAM algorithm integrated with ANN able to improve the performance of CSLAM for mobile robot using low-cost sensor. |
|---|