基于MLP和SPSO的机器人行为选择与运动控制方法

针对传统行为选择机制(ASM)不能很好地做出控制决策的问题,提出一种基于多层感知(MLP)前馈神经网络的ASM,并将其应用到移动机器人目标跟踪中。首先,根据具体应用场景预定义多个机器人行为。然后,根据机器人配备的图像和红外传感器获得的目标位置和障碍物信息,通过MLP神经网络从预定义行为中选择出所需执行的行为。另外,为了构造最优的MLP模型,采用一种简化粒子群算法(SPSO)来优化网络权值参数。机器人目标跟踪仿真的结果表明,提出的ASM能够准确选择出合适的行为,实现了控制机器人跟踪目标移动且能够避开各种障碍物。

Robot Behavior Selection and Motion Control Method Based on MLP and SPSO

For the issue that the traditional action selection mechanism (ASM) cannot make the control decision well, an ASM based on multi-layer perceptual (MLP) feed-forward neural network is proposed and applied to the mobile robot target tracking. First, multiple robot behaviors are predefined according to the specific application scenario. Then, according to the target location and obstacle information obtained by the image and infrared sensor, the MLP neural network is used to select the desired behavior from the predefined behavior. In addition, in order to construct the optimal MLP model, a simplified particle swarm optimization algorithm (SPSO) is used to optimize the network weight parameters. The simulation results of the robot target tracking show that the proposed ASM can accurately select the appropriate behavior, realize the control robot to track the target movement and be able to avoid all kinds of obstacles.