Rapid close surrounding evaluation for autonomous commercial vehicles

Abstract

A framework has been developed for local, relative navigation of an autonomous commercial vehicle in confined areas. This enables a vehicle to navigate based only on its immediate surroundings, with the vehicle as the reference. The framework uses a simulator, to simulate the environment of a mine, and a vehicle model. The latter is equipped with 2D LIDAR sensors which allow it to interpret its close surroundings. Based on this information, a navigation algorithm is selected which performs the path planning and moves the vehicle. Four such algorithms have been developed in order to handle two distinct scenarios: navigation in a corridor and navigation at an intersection. The algorithms have been evaluated for safety and accuracy and perform consistently well. The overall root mean square deviation from a reference path is less than 25% of a chosen collision threshold. Throughout the evaluation, a safe distance from all the walls was maintained. In addition, the path generation algorithms were highly efficient with an average execution time of under 2 milliseconds. This work opens up the possibilities for algorithms which can handle additional navigation scenarios and can learn and adapt to its environment.