摘要
针对强化学习算法收敛速度慢,易产生"维度灾难"的问题提出一种深度学习与强化学习相结合的算法,用于解决六足森林消防机器人的路径规划问题。采用Python方法建立二维网格地图,对复杂的环境进行模拟,减小建模的复杂性,在相同的条件下,分别对强化学习和深度强化学习算法进行仿真研究。对比仿真结果表明,深度强化学习算法下机器人到达目标点所需步长随迭代次数而减少,能使学习效率得到显著的提高,可以说明算法的收敛速度更快。
In order to solve the problem of slow convergence and dimension disaster of reinforcement learning algorithm,an algorithm combining deep learning and reinforcement learning is proposed to solve the path planning problem of six legged forest fire fighting robot.Python method is used to build a two-dimensional grid map to simulate the complex environment and reduce the complexity of modeling.Under the same conditions,the reinforcement learning algorithm and the deep reinforcement learning algorithm are simulated respectively.The simulation results show that the step length required by the robot to reach the target point under the deep reinforcement learning algorithm decreases with the number of iterations,which can significantly improve the learning efficiency and show that the convergence speed of the algorithm is faster.
作者
孙上杰
姜树海
崔嵩鹤
康玥
陈语唐
SUN Shangjie;JIANG Shuhai;CUI Songhe;KANG Yue;CHEN Yutang(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing 210037,China)
出处
《森林工程》
2020年第4期51-57,共7页
Forest Engineering
基金
国家公益性行业科研专项重大项目(201404402-03)
大学生实践创新训练计划项目(2018NFUSPITP156)。
关键词
森林消防
机器人
深度学习
路径规划
Forest fire fighting
robot
deep learning
path planning
