双足机器人跨越动态障碍物在线控制系统设计  

Design of On-line Controller for Biped Robot to Step over Dynamic Obstacle

作　　者：于薇薇[1] Sabourin C Madani K 闫杰[1] 

机构地区：[1]西北工业大学航天学院飞行控制与仿真实验室,陕西西安710072 [2]巴黎十二大LISSI实验室IUT deSenart

出　　处：《计算机测量与控制》2008年第10期1441-1443,1447,共4页Computer Measurement &Control

摘　　要：在双足机器人跨越动态障碍物的在线控制问题中,脚步规划和步态控制的学习时间是关键问题;提出了一种将机器人的步态控制和脚步规划分别独立设计的控制策略;步态控制目的是产生关节点轨迹并控制对理想轨迹的跟踪,考虑到双足机器人关节点轨迹的不连续性,应用小脑模型连接控制CMAC记忆特征步态的关节点轨迹;脚步规划的控制目标是通过对环境的视觉感知预测机器人的运动路径,算法是基于无需对动态环境精确建模的模糊Q学习算法;仿真结果表明该控制策略的可行性,并且可以有效缩短在线学习时间。In the question of on-line control for biped robot to step over dynamic obstacle, the learning time of footstep planning and gait pattern training is a crucial problem. The control strategy of design gait pattern control and footstep planning separately has been proposed. The purpose of gait pattern control is to calculate and tract the desired point trajectory. Considering the discontinuity of joint trajectories, use CMAC neural network to memorize the reference trajectories. The objective of footstep planning is to predict the locomotive path by visual obstacle detector information. The footstep planning approach is based on FQL algorithm which does not need modeling the dynamic environ- ment precisely. The simulation results present that the proposed control strategy is valid and the on-line learning time is abbreviated as well.

关 键 词：步态 脚步规划 小脑模型连接控制器 模糊Q学习算法 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]