Energy cost estimation for autonomous navigation in unstructured outdoor environments

Abstract

The navigation of autonomous mobile robots can be based on the definition of "safe points" that define a complete path. However, this method called "WayPoint based" - has been shown to be ineffective, slow and dependent on human intervention mainly in unfamiliar and unstructured environments. This occurs because there may be several variables that must be detected such as fallen trees, holes, sandy and uneven soils, among others. In addition to this method, those looking to visualize paths through LiDAR (reading via laser sensors from the environment) use related geometric analyses. An example of this is using volume as a way of identifying obstacles by deducing whether a given path is possible or not to be executed. However, in non-uniform or unstructured environments, such as forests, geometric information is often not enough to determine the transposability of obstacles. A tall grass, for example, would be identified as a geometric shape not passable by the usual methods, but it can be overcome, as it does not offer resistance to medium or large robots. In this scenario, this project seeks to calculate a transposability coefficient related to the energy cost to overcome obstacles and to develop an algorithm that, based on this coefficient and with the use of a neural network, helps in the navigation of autonomous robots in open-air and completely random environments and unknown.