Navegação segura de um carro autônomo utilizando campos vetoriais e o método da janela dinâmica

Abstract

Safe navigation is fundamental for autonomous vehicles. That requires a complete interaction with its surroundings. Self localization, motion planning, environment perception and obstacle avoidance, are important steps that must be realized by the vehicle. This work presents a safe navigation approach for a car-like robot. It uses a path planning algorithm based on Velocity Vector Fields combined together with a Dynamic Window Approach for unmodeled obstacle avoidance. Basically, the vector field is associated to a controller whose outputs are validated by the Dynamic Window Approach and applied as control inputs for the car. To assist the navigation, some known technics have been incorporated to the final solution, such as a sensor fusion system for localization and a local occupancy grid for environment perception. The methodology of this work was validated in a simulation system, where lasers and visual sensors were evaluated, and posteriorly applied to CADU (a car-like robot developed in the Federal University of Minas Gerais) that uses a stereo vision camera for obstacles detection. The results, for both cases, controlled the vehicle in an unstructured environment. The vehicle was able to track the vector field and avoid obstacles in its way. It is expected that more sensors and a better localization system would allow the car to navigate around more complex places using the methodology presented in this work.