Evaluación e implementación de un algoritmo para el diseño de rutas en ambientes dinámicos y variantes para vehículo autónomos

Abstract

This research focuses on the implementation and evaluation of algorithms for route planning in dynamic and environmental changes, applied to autonomous vehicles. MATLAB is used as a simulation tool, and two algorithms are implemented throughout the work: D* Lite and Q-learning, to compare their performance with other algorithms already implemented in the simulation environment. D* Lite is known for its ability to efficiently replan routes when there are environmental changes, while Q-learning allows continuous improvement of navigation decisions through reinforcement learning. Multiple experiments are conducted to evaluate the performance of these algorithms. Before the comparison, the logic used to develop the mentioned algorithms is explained. For the evaluation, metrics including execution time, path length, and path cost based on distance were considered. Additionally, the ability of the algorithms to avoid obstacles in controlled and randomized environments was evaluated. To validate the effectiveness of the developed algorithms, a benchmark was performed comparing these algorithms with other existing route planning methods. The results showed that both algorithms are effective in dynamic environments. D* Lite stood out for its simplicity and speed, achieving efficient and quick replanning. On the other hand, Q-learning showed great adaptability and learning ability, improving its navigation decisions as it interacted with the environment.