Multi-layer perceptron neural network mobile robot navigator in unknown environment

Indonesian Journal of Electrical Engineering and Computer Science Vol. 31, No. 2, August 2023, pp. 725~733 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v31.i2.pp725-733  725 Multi-layer perceptron neural network mobile robot navigator in unknown environment Mohammed Rabeea Hashim Al-Dahhan1,2, Ruqayah Rabeea Al-Dahhan2, Ali Tariq Radeef3 1 Department of Computer Engineering Techniques, Al-Maarif University College, Ramadi, Anbar, Iraq 2 College of Computer Science and Information Technology, Anbar University, Ramadi, Anbar, Iraq 3 Department of Computer Science, Al-Maarif University College, Ramadi, Anbar, Iraq Article Info ABSTRACT Article history: Recently, navigation in an unknown environment without hitting obstacles was considered a big challenge faced by researchers. The difficulty in finding a good mathematical model for the different systems is deciding to use artificial intelligent controllers to control the mobile robot movement. In this paper, designing two multi-layer-perceptron neural networks (MLP-NN) was done to control the movement of mobile robots in an unknown environment. The first MLP-NN is to control the linear velocity on the x-axis and angular velocity of the robot’s movement while the other MLP-NN is designed to avoid the static and dynamic obstacles faced by the robot while navigating in an unknown environment. The results show each controller's advantages in performing navigation tasks and avoiding obstacles in different environments. Received Jan 5, 2023 Revised Apr 10, 2023 Accepted Apr 16, 2023 Keywords: Mobile robot Navigation Neural network Obstacle avoidance Target seeking This is an open access article under the CC BY-SA license. Corresponding Author: Mohammed Rabeea Hashim Al-Dahhan Department of Computer Engineering Techniques, Al-Maarif University College 31000, Ramadi, Anbar, Iraq Email: 1. INTRODUCTION Initially, the term mobile robots were not used within the industrial and manufacturing field only. In the last decade started using it in the fields of medicine, rescuing, entertainment, education, agriculture, mining, military, and many more. In literature, different methods are used for safe navigation problems such as Logic Controller with Mamdani style [1] to navigate in static and dynamic environments without any collision. However, the shortest and safest path planning of a mobile robot in a static environment with different shaped obstacles by using A* is used in [2]. The path which results from this algorithm is a safe path with a large distance to obstacles but the robot cannot turn smoothly due to the sharp corner generated by the A* algorithm in the path. Fuzzy logic controller (FLC) is used in [3], [4] for finding the shortest and safest path and navigating in an unknown environment with different dynamic and static obstacles. Moreover, the type of robot used in that research has many constraints because it could not move in all directions without adjusting its rotation angle. Many researchers include the non-holonomic type of Robot in their work such as [5], [6]. Rao et al. [5] suggested neuro-fuzzy logic with integrated safe boundary algorithm to navigate safely using a non-holonomic robot. Tutuko et al. [6] and Patle et al. [7] proposed probability-based fuzzy logic for solving the path planning problems of navigation and using a non-holonomic mobile robot called “Khepera-II robot”. The result of their work was limited due to the static environment only. Rubio et al. [8] used a combination of dynamic particle swarm optimization (DPSO) with fuzzy logic (FL) to enhance the performance of the navigation for a nonholonomic mobile robot by controlling the trajectory and reducing the time traveled by the predefined target. Within the vast developments in technology, autonomous and robotics [8] many researchers are interested in Journal homepage: http://ijeecs.iaescore.com 726  ISSN: 2502-4752 autonomous driving [9], [10]. SlamThe algorithm has been used in [11] to achieve the 2D mapping, where they used an omni-robot with an X4 liDAR sensor as an indoor scanner. The main benefit of these sensors is to determine the position of the obstacle. The results show the robot can follow the walls using the proportionalintegral-derivative (PID) controller. Hani and Moin [12] slam algorithms are used to introduce the environment as static or dynamic to get prior knowledge about the environment. The results show that the algorithm is reliable for real-time applications. Navigation in different environments [13], [14], routing [15], buildings [16], industrial [17], and unmanned aerial vehicle (UAV) [18], [19]. Mobile robot has to navigate in different environments without hitting any obstacles on their route. In static environments [20], [21]. On the other hand, binary images can be used as input to navigate the environment [22], [23]. The left-hand algorithm is used in [24] to make the robot capable to follow the line in the maze map without hitting the obstacles in a static environment. On the other hand, Dagher et al. [25] also used a static environment to control the navigation of a 3WD mobile robot using a neural network. Shamsuddin et al. [26] present many algorithms for the stability of autonomous ground and unmanned vehicles. However, this work reviewed more than 50 papers talking about unmanned surface vehicle (USV) and AVG performance. Omni-directional mobile robot in a room map environment present in [27], Raspberry Pi 3 B used to connect the light detection and ranging (LiDAR) sensor with a robot to control the movement of the robot manually or automatically, as a result, the robot is capable to follow the wall in indoor room environments. Improving the robot’s movement in avoiding obstacles was done [28] using neuro-fuzzy in this work led to enhance response and decrease time response. Discovering the path in any maze map using MATLAB was presented in [29], where the robot follows the walls to find the correct path between the start point and end point. The algorithm shows good results but it has suffered from time consumption. Generating an optimal path between two points in the static environment and then minimizing the points of the path to make it smoother were demonstrated in [30]. An artificial neural network is used to control the torque required for the mobile robot to follow the path. The contribution of this work is to design and implement a controller for a mobile robot without knowing its mathematical model, then and finding the shortest path between the start and goal points on the real-environment map, also this algorithm helps the robot to avoid the static and dynamic obstacles. as a result, the robot can navigate safely in an unknown environment with less time computation by achieving the shortest path. The method will be discussed in the next part, section three will discuss the results and section four will conclude the paper's goals. 2. METHOD The most important metric in the navigation o (...truncated)