Árvores aleatórias de exploração rápida com técnicas de aprendizado de máquinas aplicadas à navegação de carros autônomos

Abstract

Mobile robotics is a promising area of engineering that combines task automation with mobility. This flexibility of movement brings a demand for methods that allow autonomous navigation of the robot, avoiding collision with any other object in space. Trajectory planning is a complex problem that often requires profound knowledge of the robot and particular heuristics to the system in which it is applied. This work proposes a systematic approach to this issue, considering the union of rapidly-exploring random trees, a classic trajectory planning algorithm, with several machine learning techniques, to overcome the difficulties associated with obtaining analytical so lutions. The proposed method is computationally evaluated in a mobile robot model similar to a passenger car, an important system considering the emergence of autonomous cars, and achieves results comparable to analytical methods, which uses broad knowledge of the system.