Navigation through unknown and uncertain environments is one of the tasks Unmanned Aerial Vehicles (UAVs) are widely used for. This is a challenging task as the environments may not be already mapped in most cases. This is where reinforcement learning comes in handy as it can enable an agent to adapt and learn in real time. Reinforcement learning works on a trial and error method. Its sequential decision‐making feature makes these algorithms more efficient to find a path in an environment with partial information. After a bunch of experimentation runs, the specialist learns the ideal strategy, which is the succession of activities that expand the complete prize. In this chapter simulations of a drone and a drone environment were made using a Robot Operating System (ROS) and Gazebo. Q‐learning algorithm and SARSA learning algorithm were used and the model was trained so that the drone was independently capable of making decisions in order to achieve the goal. The simulated results show that the drone is able to figure out an efficient path in a few iterations. The proposed work also elaborates the difference in performance of Q‐learning and SARSA learning with respect to memory‐occupation and speed of convergence.
Autonomous Navigation of Unmanned Aerial Vehicle Using Reinforcement Learning
Sustainable Mobility ; 79-103
12.12.2024
25 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Autonomous Unmanned Aerial Vehicle Navigation using Reinforcement Learning: A Systematic Review
ArXiv | 2022
|UNMANNED AERIAL VEHICLE INFRASTRUCTURE AND AUTONOMOUS NAVIGATION SYSTEM
Europäisches Patentamt | 2020
|In-furnace space autonomous navigation unmanned aerial vehicle
Europäisches Patentamt | 2022
|