【作者】 许宪东；

【导师】 关毅； 洪炳镕；

【作者基本信息】 哈尔滨工业大学 ， 计算机应用技术， 2018， 博士

【摘要】 仿人机器人进入复杂的室内环境,服务于人类的日常生活是其发展的必然趋势,定位与运动规划是机器人能够自主服务于人类的关键。目前主要研究的是简单环境下的仿人机器人定位与运动规划,而对于复杂室内环境下的定位与运动规划研究较少。虽然已存在一些轮式机器人的定位与运动规划方法,但由于仿人机器人的结构特点,导致很多适用于轮式机器人的方法并不适用于仿人机器人,仍有许多问题亟待研究解决。本文是在国家自然科学基金项目“仿人机器人同时定位和三维认知地图创建研究”的资助下,以提高仿人机器人在室内应用中所必需的自主能力为目标。针对仿人机器人复杂室内环境下定位和运动规划所面临的问题,对其涉及的若干关键技术进行了系统深入的研究。主要包括环境感知、定位和运动规划等几个方面的内容。第一,研究了仿人机器人室内地图生成问题,提出了基于混合地图的室内环境建立方法。全局层采用基于自然路标的拓扑地图。局部层针对不同的传感器,建立了二维和三维地图。采用激光测距仪建立了室内二维地图,利用RGB-D传感器构建了3D室内地图,基于QR code建立了辅助语义层。通过构建语义-拓扑-度量混合地图来建立统一构架,满足仿人机器人室内定位与导航需求。第二,研究了仿人机器人定位问题。本文针对环境地图已知和未知两种情况对仿人机器人的定位进行了研究。针对拓扑地图,提出了一种基于PNP的全局定位方法。针对度量地图,提出了一种基于KLD-蒙特卡罗的仿人机器人定位方法。针对环境地图未知情况,提出了一种基于滑动窗口扩展卡尔曼滤波的仿人机器人定位方法。这三种方法可根据需要应用于仿人机器人的全局与局部定位,获得较好的定位效果。第三,针对仿人机器人室内运动需面对楼梯和斜坡等障碍物的问题,研究了仿人机器人爬楼梯、斜坡问题。提出了一种基于改进NSGA-II的仿人机器人上楼梯和斜坡的运动规划方法。建立了仿人机器人七连杆模型,设计了上楼梯、斜坡运动模型。通过引入多目标优化方法,实现仿人机器人上楼梯、斜坡运动,可以较好的满足多目标的需求。第四,仿人机器人完成复杂任务一般需要通过全身运动规划完成。针对仿人机器人全身运动规划问题,提出了一种基于运动捕捉与动量分解的分阶段全身运动生成方法。离线阶段通过B样条拟合关节角度,并通过优化获得轨迹。在线阶段通过采用模型预测与分解动量控制相结合,完成仿人机器人在线运动控制。实验表明,该方法可以有效的实现仿人机器人全身运动规划,并可在全身运动过程中保持较好的稳定性。更多还原

【Abstract】 It is the inevitable trend that a humanoid robot enters the complex indoor environment,and serves the daily life of human being.To serve the people autonomously,location and motion planning are essential technologies for a humanoid robot.Current research is focused on localization and motion planning in a simple environment,while few of them consider those two problems in complex indoor environment.Although there are some localization and motion planning methods for wheeled robots,due to the special characteristics of humanoid robot,most methods for wheeled robot are not suitable for humanoid robot,and there are still many problems to be solved.Supported by National Natural Science Foundation of China “Research on simultaneous localization and 3D cognitive map-building for a humanoid robot”,the purpose of this research is to improve the autonomous ability of humanoid robot.It mainly concerns about the problems of localization and motion planning for humanoid robot in complex indoor environment.Key technologies involved are systematically and deeply studied.It mainly includes environment perception,location and motion planning.Firstly,the problem of indoor mapping for a humanoid robot is studied.It proposed a hybrid map method for mapping an indoor environment.Global topological map is constructed by natural landmarks,and 2D and 3D maps are built by different sensors for local metric layer.The 2D map is established by an laser rangefinder.And the 3D map is established by a RGB-D camera.In addition,the auxiliary semantic layer is established based on QR code.And a unified framework of semantic-topological-metric hybrid map is set up,which can meet the requirements of indoor positioning and navigation for humanoid robots.Secondly,the localization problem of a humanoid robot is studied.In this paper,the localization problem is deeply studied in two cases: the environment map is known and unknown.A global positioning method based on PNP is proposed for topological map,and a positioning method based on KLD-Monte Carlo is presented for local metric map.To deal with the unknown environment,a method named Sliding Window Extended Kalman Filter is proposed.Those three methods can be applied to global and local localization for humanoid robot,which can obtain better positioning results.Thirdly,the problem of humanoid robot climbing stairs and slopes is studied.A motion planning method based on the improved NSGA-II is proposed.A seven-link model of humanoid robot is established,and motion model of climbing stairs and slopes is set up.By introducing the multi-objective optimization method,motion of climbing stairs and slope are realized,which can meet the multi-objective demand.Fourthly,in order to accomplish a complex task,whole body motion planning is necessary.To aim at this problem,a new method of staged whole-body motion planning based on motion capture and resolved momentum control is proposed.In the offline stage,the B spline is used to fit the joint angle and the trajectory is obtained by optimization.In the online stage,the motion control of humanoid robot is realized by combining Model Predictive Control with Resolved Momentum Control.Experimental results show that this method can effectively realize the whole body motion planning for a humanoid robot,and the robot can keep stability during the whole body movement.更多还原