基于深度强化学习的AGV智能导航系统设计

Design of AGV intelligent navigation system based on deep reinforcement learning

针对现有的AGV在大规模未知复杂环境中进行自主导航配送的问题,基于深度强化学习完成了AGV智能导航系统设计。首先,结合传感器对周围的障碍物进行探测感知,利用DDPG(deep deterministic policy gradient)算法实现AGV小车从环境的感知输入到动作的直接输出控制,帮助AGV完成自主导航和避障任务。此外,针对训练样本易受环境干扰的问题,提出了一种新颖的DL(disturb learning)-DDPG算法,通过对学习样本中相关数据进行高斯噪声预处理,帮助智能体适应噪声状态下的训练环境,提升了AGV在真实环境中的鲁棒性。仿真实验表明,经改进后的DL-DDPG算法能够为AGV导航系统提供更高效的在线决策能力,使AGV小车完成自主导航与智能控制。

Aiming at autonomous navigation and delivery of AGV in large-scale complicated and unknown environment,this paper put forward an autonomous online decision-making algorithm based on deep reinforcement learning.Specifically,combining with sensors to detect and perceive surrounding obstacles,the method used DDPG algorithm to realize the input of environmental perception and action direct output control,which helped the AGV complete autonomous navigation and autonomous obstacle avoidance tasks.To solve the problem,it disturbed the training samples easily by the environment,the algorithm preprocessed the relevant data with Gaussian noise in the learning sample,which helped the agent adapt to the training environment under noise and improve its robustness in real environment.Simulation results show that the improved DL-DDPG algorithm can provide more efficient online decision-making ability for the control system and enable the competency of autonomous navigation and intelligent control of AGV.