Q-Learning based Local Path Planning for UAVs with Different Priorities

• Kevin B. de Carvalho UFV
• O. B. Hiago Batista UFV
• Leonardo A. Fagundes-Júnior UFV
• Alexandre S. Brandão UFV

Resumo

Path planning is a crucial component of autonomous navigation and frequently demands different priorities such as path length, safety, or energy consumption, with the latter being particularly important in the context of unmanned autonomous vehicles. In many applications, the agent may have to react to environment shifting. Algorithms such as geometric and dynamic programming as well as techniques such as artificial potential fields have been employed to tackle this local planning problem. In recent years, machine learning has gained more evidence in many research fields due to its flexibility and generalization capabilities. In this study, we propose a Q-learning-based approach to local planning, which weighs three crucial factors- path length, safety, and energy consumption- that can be freely adjusted by the user to suit its application’s needs. The performance of the proposed method was tested in simulated static and dynamic scenarios as well as benchmarked with a baseline approach. The results show that it can perform well in both kinds of environments without struggling with the commom pitfalls of other local planning algorithms.

Palavras-chave: Mobile Robotics, Reinforcement Learning, Local Path Planning, Unmanned Aerial Vehicles