Contributions to LiDAR based SLAM and Computational Ethology.

Abstract

This Thesis deals with two different topics centered about applications of Computational Intelligence techniques. The first topic is the implementation of simultaneous localization and mapping (SLAM) algorithms that are appropriate for low-cost LiDAR sensors, specifically the Quanergy M8. Conventional and Deep Learning algorithms have shown shortcomings dealing with this data, hence this Thesis proposes a novel hybrid SLAM algorithm that achieves good results over in-house datasets captured with the low cost LiDAR sensor. The second topic tackled in this Thesis is the discrimination of animal models on the basis of pressure signals. For this task, we work on real experimental data provided by a collaborating neurosciences team. The Thesis deals with the selection of signal features and the experimentation with a diversity of state of the art machine learning algorithms. The application of transfer deep learning upon signal spectrogram images improves significantly over conventional machine learning algorithms, concluding that it is feasible to discriminate animal models on the basis of pressure signal captured during locomotion periods.