Recently service robots are increasingly been deployed for different applications such as for homes, hospitals, and other service industries. For efficient navigation of mobile robots in complex indoor and outdoor environments, it is essential for the robot to perform collision-free and optimized planning in dynamic scenarios. Recently, reinforcement learning methods, a type of machine learning technology are used for the navigation tasks in mobile robots and is actively been researched. It has several advantages that the robot can learn and adapt to new and unknown environments quickly by training over several iterations of different scenarios. In this paper, we propose a navigation policy for a mobile robot equipped with a 2D-LiDAR sensor based on the Proximal Policy Optimization (PPO) of a stochastic approach. Our method also includes a stochastic operation which reduces the computational cost of a training session. We also trained our robot in multiple virtual environments to speed up the learning model ’s training. For this research, the navigation test was executed only in a virtual simulator.