Scholarlink

<p>In this paper, we use reinforcement learning (RL) to address the motion planning problem of a quadrotor unmanned aerial vehicle (UAV) navigating an unknown environment. The quadrotor UAV is equipped with a vision sensor that collects obstacle information during the flight, and feedback linearization and linear quadratic tracker are used for controller design. The obtained obstacle data points construct a spherical bounding box, efficiently storing vision data, and establish an observation space of the RL framework. Additionally, an action space consists of a velocity command which is fed into the control system of the quadrotor UAV. A reward function is designed for the quadrotor UAV to avoid an obstacle and reach a goal point. Numerical simulation is conducted to train the quadrotor and Monte Carlo simulation is performed to verify the robustness of the proposed algorithm. The trained quadrotor UAV can deal with multiple obstacles and the result is compared to a collision cone approach.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Motion Planning of Quadrotor UAV via Reinforcement Learning in Unknown Obstacle Environment

  • 期刊论文

  • 发表日期：

    2025年10月16日

查看全文
  • Jongho Park,
  • Suwon Lee,
  • Junsoo Lee

摘要

In this paper, we use reinforcement learning (RL) to address the motion planning problem of a quadrotor unmanned aerial vehicle (UAV) navigating an unknown environment. The quadrotor UAV is equipped with a vision sensor that collects obstacle information during the flight, and feedback linearization and linear quadratic tracker are used for controller design. The obtained obstacle data points construct a spherical bounding box, efficiently storing vision data, and establish an observation space of the RL framework. Additionally, an action space consists of a velocity command which is fed into the control system of the quadrotor UAV. A reward function is designed for the quadrotor UAV to avoid an obstacle and reach a goal point. Numerical simulation is conducted to train the quadrotor and Monte Carlo simulation is performed to verify the robustness of the proposed algorithm. The trained quadrotor UAV can deal with multiple obstacles and the result is compared to a collision cone approach.