基于视觉与行为模型的机器人目标跟踪

Robot target tracking based on vision-behavior model

由于存在通信延时等问题,月球机器人必须具备一定智能来进行行为控制。Brooks提出机器人的包容结构理论重点在于强调机器人不同控制层间的联系,以及机器人不同行为功能的分配。本文提出一种视觉与行为模型,主要描述在同一控制层内视觉传感器与行为控制之间的关系,它将机器人的视觉行为与运动行为紧密联系起来,在运动的同时计算地面光流场变化来学习当前机器人动作状态,利用颜色信息的多窗口目标跟踪获取目标属性,最后采用强化学习方法,规划机器人动作,提高了运动控制的稳定度和精确度。

The lunar rover should be equipped with some intelligence to carry out the tasks independently because of the time delay. Prof. Brooks has brought forward layered control system for the mobile robot, which is focused on decomposing the whole tasks into several concurrent behaviors in layers and constructing the relationship between the different layers. In this paper, a vision-behavior model was brought up, in which the vision sensor and behavior controller were closely tied up. While the robot moving, the optical flow of the floor is gain to analyze the robot motion state. Muti-Windows are used in target tracking by means of color information. And then the reinforcement learning is used to direct the robot behavior. This method was proved to be stable and accurate.