Occupancy grid map algorithm with neural network using array of infrared sensors

Abstract

Occupancy grid map is a map representation that shows the occupancy of spaces, whether there is any object in a particular area or it is a free space. This map representation is also commonly known as a grid map. However, the accuracy of the occupancy grid map is highly dependent on the accuracy of the sensors. In this paper, low cost and noisy sensors such as infrared sensors were used with the occupancy grid map algorithm integrated with a neural network. The neural network was used to interpret adjacent sensor measurements into cell’s occupancy value in the grid map. From the simulation experiments, it is observed that, that neural network-integrated algorithm has a better map estimate throughout robot’s navigation with mean of 28% more accurate compared to occupancy grid map algorithm without neural network. This finding is beneficial for implementation with simultaneous localization and mapping or commonly known as SLAM problem. This is because SLAM algorithm makes use of both estimations of environment’s map and robot’s state. Thus, a better map estimate throughout the robot’s journey can improve a robot’s state estimate as well.

The Effect of Artificial Neural Network Towards the Number of Particles of Rao-Blackwellized Particle Filter using Laser Distance Sensor

“…However, higher number of particles might suffer from forbidding memory burden and higher computational cost. This problem can be overcome by integrating the SLAM technique with an artificial neural network (ANN) while using low-cost sensors [5], [19], [20], [33]. The noisy dataset from the sensor of the mobile robot are used to train the ANN learner.…”

“…In this paper, there are two strategies of dataset that have been reviewed to train the ANN network. Firstly, the training network using the position of each of the grid cells of OGM [5], [19], [33]. Secondly, by using the distance from sensor to obstacles [20].…”

“…This ANN is trained to estimate the occupancy probability value x, y position of the cell. The same architecture of ANN is used in [5], [33]. The authors used the front four sensors" measurements of Khepera III robot that are closest to the cell position as the input of ANN.…”

Rao-Blackwellized Particle Filter with Neural Network using Low-Cost Range Sensor in Indoor Environment

“…The cell's occupancy value is then converted into log odd notation in the occupancy grid map algorithm which is part of the map update step. The overall map update step implemented is described in our previous work [19]. To interpret cell's occupancy, selected sensor measurements with encoded cell's position were used as the neural network inputs.…”

“…While for robot's state accuracy, the root mean squared (RMSE) value was used as performance measure. This analysis methods is described in our previous work [19].…”