机械工程学报
機械工程學報
궤계공정학보
CHINESE JOURNAL OF MECHANICAL ENGINEERING
2015年
7期
1-22,23
,共23页
BigDog四足机器人%运动控制地形还原%虚拟模型%自主性%智能性%LS3机器人%阿特拉斯机器人
BigDog四足機器人%運動控製地形還原%虛擬模型%自主性%智能性%LS3機器人%阿特拉斯機器人
BigDog사족궤기인%운동공제지형환원%허의모형%자주성%지능성%LS3궤기인%아특랍사궤기인
BigDog quadruped robot%ground plane estimation of locomotion control%virtual model%autonomy%intelligence%LS3 robot%Atlas robot*
对BigDog四足机器人的核心技术进行分析,适应复杂地形是BigDog的设计主线。提高横、纵自由度联动能力是BigDog结构设计主要突破点。机体重心颠簸起伏、机体重心自扰动等不良运动特性是四足机器人控制难度大的主要原因。液压动力系统的构成和优点将被剖析,解决腿类移动装置的驱动问题是液压系统研发的根本目的。支撑腿打滑及俯仰和横滚角度是否过大作为监测机体运动安全状态的参数。惯导和关节编码器可检测机身与肢体的状态,借助压力传感器可还原落足点地形,三者合一可构建虚拟模型。借助虚拟模型可求算机体重心等关键控制处理中间参数,运动控制系统可实施粗略的动作预演及精确的运动学和动力学规划。规划模型与样机模型的偏差作为反馈值实施闭环控制。建立以三维激光扫描仪和双目视觉为主的导航系统,视觉地形还原功能可帮助 LS3安全跨越岩石地形,软件系统将各种基本功能整合为有机的整体。机器人的自主性与智能性被讨论,利用BigDog/LS3与好奇号火星探测器作对比并加以分析。BigDog目前存在的几个主要问题:液压系统无法瞬时大幅增压、机械传动各种损伤、仿生设计的不彻底性。LS3机器人针对BigDog的不足,多个改进环节被分析。猎豹、野猫、Petman 等机器人被简要分析。阿特拉斯双足机器人借助虚拟模型可实现机械臂碰撞保护功能,遭受外力撞击可迅速恢复平衡状态。
對BigDog四足機器人的覈心技術進行分析,適應複雜地形是BigDog的設計主線。提高橫、縱自由度聯動能力是BigDog結構設計主要突破點。機體重心顛簸起伏、機體重心自擾動等不良運動特性是四足機器人控製難度大的主要原因。液壓動力繫統的構成和優點將被剖析,解決腿類移動裝置的驅動問題是液壓繫統研髮的根本目的。支撐腿打滑及俯仰和橫滾角度是否過大作為鑑測機體運動安全狀態的參數。慣導和關節編碼器可檢測機身與肢體的狀態,藉助壓力傳感器可還原落足點地形,三者閤一可構建虛擬模型。藉助虛擬模型可求算機體重心等關鍵控製處理中間參數,運動控製繫統可實施粗略的動作預縯及精確的運動學和動力學規劃。規劃模型與樣機模型的偏差作為反饋值實施閉環控製。建立以三維激光掃描儀和雙目視覺為主的導航繫統,視覺地形還原功能可幫助 LS3安全跨越巖石地形,軟件繫統將各種基本功能整閤為有機的整體。機器人的自主性與智能性被討論,利用BigDog/LS3與好奇號火星探測器作對比併加以分析。BigDog目前存在的幾箇主要問題:液壓繫統無法瞬時大幅增壓、機械傳動各種損傷、倣生設計的不徹底性。LS3機器人針對BigDog的不足,多箇改進環節被分析。獵豹、野貓、Petman 等機器人被簡要分析。阿特拉斯雙足機器人藉助虛擬模型可實現機械臂踫撞保護功能,遭受外力撞擊可迅速恢複平衡狀態。
대BigDog사족궤기인적핵심기술진행분석,괄응복잡지형시BigDog적설계주선。제고횡、종자유도련동능력시BigDog결구설계주요돌파점。궤체중심전파기복、궤체중심자우동등불량운동특성시사족궤기인공제난도대적주요원인。액압동력계통적구성화우점장피부석,해결퇴류이동장치적구동문제시액압계통연발적근본목적。지탱퇴타활급부앙화횡곤각도시부과대작위감측궤체운동안전상태적삼수。관도화관절편마기가검측궤신여지체적상태,차조압력전감기가환원락족점지형,삼자합일가구건허의모형。차조허의모형가구산궤체중심등관건공제처리중간삼수,운동공제계통가실시조략적동작예연급정학적운동학화동역학규화。규화모형여양궤모형적편차작위반궤치실시폐배공제。건립이삼유격광소묘의화쌍목시각위주적도항계통,시각지형환원공능가방조 LS3안전과월암석지형,연건계통장각충기본공능정합위유궤적정체。궤기인적자주성여지능성피토론,이용BigDog/LS3여호기호화성탐측기작대비병가이분석。BigDog목전존재적궤개주요문제:액압계통무법순시대폭증압、궤계전동각충손상、방생설계적불철저성。LS3궤기인침대BigDog적불족,다개개진배절피분석。작표、야묘、Petman 등궤기인피간요분석。아특랍사쌍족궤기인차조허의모형가실현궤계비팽당보호공능,조수외력당격가신속회복평형상태。
The core technology of the BigDog quadruped robot is analyzed. Adapting to the rough terrain is the main design clues of the BigDog. Improving horizontal and vertical degrees of freedom linkage ability is the main innovation of structure design. Not good motion characteristics, such as robot’s center of gravity ups and downs and self disturbance are the main reasons for being difficult to control. The components and advantage of the hydraulic power system are analyzed. Solving the driver problem of legged vehicles is the fundamental goal of the hydraulic system development. Supporting leg slipping or not, pitch and roll angle of the body too large or not are the main parameters as monitoring robot’s movement condition. IMU and joint encoder can detect the state parameters of the body and limbs. Terrain of foot placement can be restored by pressure sensor. Three-in-one can build a virtual model. By the virtual model, robot’s center of gravity and other key control process parameters can be calculated. At the same time, locomotion control system can do action drill roughly and accurate planning of kinematics or dynamics. The deviation of planning and prototype model is taken as the feedback for closed-loop control. LS3 constructs the navigation system of three-dimensional laser scanner and binocular vision as the main. LS3 can stride across rocky terrain by visual terrain reconstruction. Software system can integrate all the basic functions as an organic whole. Autonomy and intelligence of robot are discussed. BigDog/LS3 and Curiosity Mars Rover are compared and analyzed. BigDog has three big problems currently: instantaneously unable to increase hydraulic value significantly, all kinds of damage in mechanical transmission, bionic design not thoroughness. For the inadequacies of BigDog, several improvements are analyzed on the LS3. Petman, Cheetah and Wildcat robot are briefly analyzed. Atlas biped robot has crash protection function and can recovery equilibrium status quickly after external force hitting by virtual model.