Indoor localization has been an active research area for the last two decades. This emerged in the context of providing a mobile robot the capability to conduct navigation tasks in indoor environments. Although the sensing technologies and techniques proposed for indoor robot localization have proven to be reliable solutions, these cannot be adopted as a solution to people or object localization for indoor environments, particularly, due to their high computational cost and power requirements. In order to mitigate these issues, a low-power consumption sensing technology, based on the strength of WiFi signals, is being studied. Nevertheless, a concern when working with these signals is their vulnerability to interference. This paper exploits the use of machine learning is two different architectures for localization and present how a data filtering technique can alleviate interferences. A step into a fault tolerance approach is also given, presenting that the system can maintain certain reliability even losing some of its parts.

The world population growth poses several challenges, including the increase need of food supply. To overcome this issue, large-scale agricultural production plays an essential role. Besides, agriculture has become a strategic area in the economy of several countries. In this scenario, a common practice is the production management with the use of plant protection products in order to protect the crop from the harmful action of agricultural pests. During this activity, the applied products commonly drifts out of the target region, contaminating other environments or causing excessive deposition (due to product overlapping). The possibility of an spraying system that adapts to the weather conditions can reduce the off drift, increasing the accuracy of the deposition and, consequently, providing a more suitable environment for the growing crop. For such, the spraying system should perform the prediction of the deposition at run time, which is hard to achieve due to the high computational cost of current approaches. This article proposes and experimentally evaluates a new approach based on Machine Learning that predicts the amount of pulverized products with lower computational cost and high accuracy. Experimental results show that the proposed approach represents well the actual deposition. For the prediction of product deposition, it was able to obtain mean absolute error of 0.096 μL/cm ² when employing Artificial Neural Networks and 0.089 μL/cm ² when employing Regression Trees. These results stimulate new studies to compare the proposed approach with existing solutions.