附录1机械英语文章翻译_机械英语文章翻译9

附录1机械英语文章翻译_机械英语文章翻译9 附录1 Distributed Control of Gait for a Humanoid Robot Gordon Wyeth and Damien Kee School of Information Technology and Electrical Engineering University of Queensland, St. Lucia, Queensland, 4072, Australia {wyeth,damien}@itee.uq.edu.au Abstract. This paper describes a walking gait for a humanoid robot with a distributed control system. The motion for the robot is calculated in real time on a central controller, and sent over CAN bus to the distributed control system. The distributed control system loosely follows the motion patterns from the central controller, while also acting to maintain stability and balance. There is no global feedback control system; the system maintains its balance by the interaction between central gait and "soft" control of the actuators. The paper illustrates a straight line walking gait and shows the interaction between gait generation and the control system. The analysis of the data shows that successful walking can be achieved without maintaining strict local joint control, and without explicit global balance coordination. 1 Introduction Humanoid robots typically require coordinated control of a large number of joints. In most existing implementations of humanoid robots, coordination is achieved by the use of a central control computer that interfaces to all sensors and actuators providing local control of joint positions and torques as well as global control of balance and posture. This paper describes a distributed approach to control and coordination that provides local control of position and torque at each joint in a fashion that maintains global balance and posture. 1.1 Paper Overview After a brief description of related work, the paper describes the GuRoo robot that forms the basis for the later experiments. Details of the architecture and design of the robot are followed by a description of the computing system that supports the distributed control system. The paper then describes the approach to distributed control and provides details of gait generation, including results gathered from a straight line walk. 2 Related Work The OpenPino project [Yamasaki, 2000], utilizes a very centralized approach to humanoid control. An onboard SH2 micro-controller is responsible for taking high level commands from a PC, such as walk forward, stop etc, and converts them into positioning formation. These positions are converted into PWM signals by a single CLPD, responsible for controlling all 26 degrees of freedom. The University of Waseda's humanoid, WABIAN, employs a similar control sys-tem, with high level commands generated by the onboard Pentium 166Mhz computer[Yamaguchi, 1998].The resulting velocity profiles are fed into one of two 16 channel D/A boards via an ISA bus, which supply the motor drivers the required signals to actuate the motors. This system is physically centralized with all computational equipment and motor drivers located in the torso. Similarly, H6 from the University of Tokyo makes use of an onboard PIII 700Mhz computer to generate high level commands [Nishiwaki, 2000]. A pair of Fujitsu I/O controller, similar to WABIAN's, generates the control signals necessary for the motors. Individual motor drivers supplying the necessary power are located physically close to each motor. 2.1 The GuRoo Project GuRoo is a 1.2 m tall, fully autonomous humanoid robot designed and built in the University of Queensland Robotics Laboratory [Wyeth, 2001]. The robot has a total mass of 34 kg, including on-board power and computation. GuRoo is currently capable of a number of demonstration tasks including balancing, walking, turning, crouching, shaking hands and waving. The robot has performed live demonstrations of combinations of these tasks at various robot displays. The intended challenge task for the robot is to play a game of soccer with or against human players or other humanoid robots. To complete this challenge, the robot must be able to move freely on its two legs. Clearly, the robot must operate in a completely autonomous fashion without support harnesses or wiring tethers. The current GuRoo robot cannot withstand the impacts associated with playing soccer, but serves as an excellent platform for research into the design of balance behaviors and dynamic gait control. The location and axis of actuation of each joint can be seen in Figure 1. Fig. 1. The GuRoo Humanoid robot and the degrees of freedom of each joint. 2.2 Electro-Mechanical Design The key element in driving the mechanical design has been the choice of actuator. The robot has 23 joints in total. The legs and abdomen contain 15 joints that are required to produce significant mechanical power, most generally with large torques and relatively low speeds. The other 8 joints drive the head and neck assembly, and the arms with significantly less torque and speed requirements. The 15 high power joints all use the same motor-gearbox combination consisting of a Maxon RE 36 motor with a gearbox reduction of 156:1. The maximum continuous generated output torque is 10 Nm. Each motor is fitted with an optical encoder for position and velocity feedback. The 8 low power joints are Hi-Tec RC servo motors model HS705-MG. with rated output torque to 1.4 Nm. The motors that drive the roll axis of the hip joints are supplemented by springs with a spring constant of 1 Nm/degree. These springs serve to counteract the natural tendency of the legs to collide, and help to generate the swaying motion that is critical to the success of the walking gait. Power is provided by 2 × 1.5Ah 42V NiCd packs for the high power motors, and 2 x 3Ah 7.2 V NiCd battery packs for computing and servo operation. The packs are chosen to give 20 minutes of continuous operation. 2.3 Sensing The position feedback from the encoders on the high power joints provides 867 encoder counts per degree of joint motion. In addition, each DSP can measure the current to each motor. Provision has also been made for inertial and balance sensors, as well as contact switches in the feet and in the joints. 3 Distributed Control Network A distributed control network controls the robot, with a central computing hub that sets the goals for the robot, processes the sensor information, and provides coordination targets for the joints. The joints have their own control processors that act in groups to maintain global stability, while also operating individually to provide local motor control. The distributed system is connected by a CAN network. 3.1 Central Control The central control of the robot derives the joint velocities required to perform the walking gait. A PIII 1.1GHz laptop currently calculates velocities for all 23 degrees of freedom in real time. The velocities are passed along a serial link to a distribution board which serves as a bridge between the serial bus and Control Area Network(CAN) Provision has been made to port this control to a Compaq IPAQ mounted on the robot to enable true autonomy, free of any tethers. 3.2 Joint Controllers All joint controllers are implemented using a TMS320F243 Digital Signal Processor from Texas Instruments, a 16 bit DSP designed for motor control. The availability of the CAN module in this series, along with boot loader programmable internal Flash memory makes the device particularly attractive for this application. Five controller boards control the 15 high power motors, each board controlling three motors. A sixth controller board controls the eight RC servo motors. Figure 2 outlines the interaction between the various node on the control network. 4 Software The software consists of four main entities: the global movement generation code, the local motor control, the low-level code of the robot, and the simulator. The software is organized to provide a standard interface to both the low-level code on the robot and the simulator. This means that the software developed in simulation can be simply recompiled to operate on the real robot. Consequently, the robot employs a number of standard interface calls that are used for both the robot and the simulator including reading the encoders and setting PWM values as well as the transfer of CAN packets. 4.1 Gait Generation The gait generation module is responsible for producing realizable trajectories for the 23 joints so that the robot can perform basic behaviors, such as standing on one leg, crouching and walking. The most important properties of the trajectories are that they are smooth and that they can be linked together in a smooth fashion. Smoothness of motion implies less disturbance to the control of other joints. Based on these criteria, a normalized joint movement as shown in Figure 3 is applied to all motor trajectories. The trajectories are generated from a parameterized sinusoidal curve where is the desired joint velocity, is the total joint angle to move and T is the period of that movement. The trajectory contrasts with typical trajectories generated for robotic manipulators, which typically focus on smoothness of the end effector motion rather than smoothness at the joint. Trajectories for each of the motors may be coordinated by using the same beginning time for the motion and specifying the same period for the trajectory. Trajecto- ries may be naturally linked as the velocities all reach zero at the beginning and the end of a motion. Section 5 will illustrate how trajectories may be coordinated and linked to perform a walking operation. Fig. 3. The trajectory used for a total joint movement of 1 radian over a period of 1 second. 4.2 Joint Controller Software There are two types of joint controller boards used in the robot - five controlle boards control the fifteen high power motors and one controller controls the eight lopower motors. The controller software for the low power motors is a single interrupt routine that is triggered by the arrival of a CAN packet addressed to the controller'mailbox. The routine reads the CAN mailbox for the change in position sent by th gait generation routine. The PWM duty cycle that controls the position of the Rservos is varied accordingly. The control loop for the high power controllers has two interrupt routines. As fo the low power controller, an interrupt is executed upon receipt of trajectory data in the CAN mailbox. The data is used to set the velocity setpoints for the motor controroutine. There is also a periodic interrupt every 500 s to run the motor controlsof ware. The motor control routine compares the error between velocity setpoint and thencoder reading and generates a PWM value for the motor based on a ProportionaIntegral control law. The routine also checks the motor current against the currenlimits, and adjusts the PWM value to prevent over-current situations. The Pcontrol law on each joint has been hand tuned to provide both good trajec tory following for typical velocity input profiles, and spring-damper model impedanc to torque disturbances from gravity and the cross-coupling torques from other joint The "soft" response of this control law to disturbance prevents torques being transmited throughout the robot and helps to maintain global stability. The disadvantage ithat the controller suffers from position error that must be accounted in the gait generation software. Section 5 illustrates how this potential liability is turned to an ass in the generation of a dynamically stable walk. 4.3 Low-Level Code The lowest level of code on the robot provides direct access to the sensors and communication system. The level of abstraction provided by function calls at this levelaids in the cross development of code between the simulator and the real robot. 4.4 Simulator The simulator is based on the DynaMechs project [McMillan, 1995], with additions to simulate specific features of the robot such as the DC motors and motor drives, the RC servos, the sensors, the heterogeneous processing environment and the CAN ne-work. These additions provide the same interface for the dynamic graphicalsimula-tion as for the joint controller and gait generation code. Theparameters for the simulator are derived from the CAD models and the data sheets from known components. These parameters include the modified Denavit-Hartenberg parameters that describe the robot topology, the tensor matrices of the links and the various motor and gearbox characteristics associated with each joint. The surface data from the CAD model is also imported to the simulator for the graphical display. For the high power DC motor joints, the simulator provides the programmer with readings from the encoders and the current sensors, based on the velocities and torques from the dynamic equations. In the case of the RC servos, the simulator up-dates the position of the joints based on a PD model with a limited slew rate. The programmermust supply the simulator with PWM values for the motors to provide the control. The simulator provides fake interrupts to simulate the real events that are the basis of the control software. The simulator uses an integration step size of 500 s and updates the graphical display every 5ms of simulated time. When running on 1.5 GHz Pentium 4 under Windows 2000, the simulation updates all 23 joints at a very useable 40% of real time speed. 5 Walking The robot can walk with a step rate of 1 Hz using a step length of 100 mm. The walk is open-loop; there is no feedback from the joint controllers to the gait generation software. The lack of global feedback, combined with the absence of a global balance sensor presents a substantial challenge in walking algorithm design. 5.1 Walking Algorithm The robot uses a simplified version of a typical human gait. In particular, it limits the swing of the legs to prevent balance disturbance as this cannot be corrected without global balance control. In order to minimize the accelerations of the torso, head and arms (which make up 1/3 of the mass of the robot), the robot maintains a constant relative position of the torso, such that the face of the torso is always normal to the direction of travel. The allowable roll of the torso is also limited. The stabilization of the torso also reduces disturbances from gravity to the control of the leg joints. Before walking, the robot loads each motor against gravity by performing a slight squat that introduces a 6 degree ankle pitch, with the knee and hip pitch joints set to keep the torso upright. The initial loading of the joints reduces the likelihood of backlash in the gearheads. The walking gait commences with a side-to-side sway generated from the roll axes of the ankles and hips. The sway frequency of 0.5 Hz is sympathetic with the spring mass system formed by the ankle controllers with the mass of robot. The sway sets up the pattern of weight transfer from one foot to the other necessary to swing the legs alternately to achieve walking. At each extreme of the sway, the inertia of the upper body ensures the ZMP (Zero Moment Point) of the robot lies within the support polygon formed by the support foot, even thought the centre of mass may not. This action places requires less torque from the hip and ankle roll actuators, as the motion due to gravity brings the robot away from the extreme of each sway. Fig. 4. Frontal view of the walking process. Once the sway is sufficient to leave no ground reaction force on the non-supporting foot, the non-supporting leg is lifted using the pitch axes of the hip, knee and ankle. Between the lifting and lowering of the non-supporting leg, each yaw axis motor twists, so that the non-supporting leg swings forward to create the step. When the swing leg contacts the ground, the robot is dynamically stable, with the centre of mass over the supporting foot in the frontal plane, but in front of the toes in the sagittal plane. The robot then swings across to the other foot, repeating the sequence and progressing with the walk. 5.2 Analysis of Results The motion of the robot is best analyzed by comparing the desired velocity from the gait generation module to the actual velocity at each joint. Figure 5 shows this comparison for the motion of the hip, knee and ankle in the roll, pitch and yaw axis. The graphs are initialized midway through the double support phase, with both legs in contact with the ground. The graphs comprise one second of data, describing the right leg as it moves from the double support phase, through the swing phase back to the double support phase. At the point t = 0.25 s, the swing leg starts to lift and loses contact with the ground. With the hip roll axis of the swing leg no longer contributing to the support of the robot, the spring located in this axis briefly dominates the actual velocity causing the overdamped oscillation seen at hip joint at this time. Once the foot leaves the ground the ankle roll motor switches from driving the leg from the foot,to driving the foot from the leg. This large decrease in relative inertia results in a brief increase in the magnitude of the ankle roll velocity. The foot has a relatively low inertia compared with the rest of the robot, and as such the PI controller has little trouble following the desired velocity until the foot again makes contact with the ground. The robot reaches the extreme of each sway at t = 0.5 s, where all motion in the roll plane ceases. The swing leg is now theoretically fully lifted, although the knee and hip pitch do not reach their desired positions until T=0.6s. Fig. 5. Desired vs Actual joint angles for straight line walking over a 1 second period. Graph start during the double support phase and follow the right leg through the swing phase.The actual joint velocity profile for each pitch motor in the swing leg shows the increasing effect of gravity and leg inertia through the swing stage. The integral term in the PI controller seeks to eliminate steady state error, and as such, dominates the actual velocity, driving each pitch motor to its desired position. As the motor does not reach the maximum desired velocity, it is forced to lengthen the movement time to ensure the areas under each graph are equal. The low proportional term results in the poor tracking of the desired velocity, but enables the joint to better deal with external disturbances. The comparison of the actual velocity with the desired velocity for each pitch motor in each leg degrades from the ankle to the knee to the hip. The ankle need only accelerate the foot, whereas the knee must accelerate the foot and lower leg. The hip pitch must accelerate the entire leg during the swing phase. The motion of the yaw axis as the swing leg is lifted, propels the robot forward. When the yaw motion occurs on the support leg, the momentum of the robot causes the joint to overshoot its position. The swing leg is then lowered placing the robot into a double support phase. The friction of two feet against the ground and the weight of the robot on the support leg prevents the yaw axis positional error from being re-solved. This provides a pre-loading of the joint that supports the motion of the next of yaw swing phase. As the robot sways back to the other side, weight is gradually released from the supporting foot, until the torque acting on the joint overcomes the co-efficient of friction between the foot and the floor. This is not necessarily the point at which the swing leg loses contact with the ground. By time the leg has resolved this error, the joint is experiencing the yaw motion associated with the swing leg twist. As a result, the area under the curve for the hip yaw during the swing phase is greater than the desired area. Contact with the ground is achieved at T=0.85s and once made, the robot returns to the double support phase of its gait. Both the hip and ankle of the swing leg now assist the support leg roll motors to sway the robot across to the other foot, and in the process gradually switch the roles of the support and swing leg 6 Conclusions This paper has illustrated that a humanoid robot can walk without the need for explicit global feedback, or tightly controlled joint trajectories. By combining a group of loosely coordinated control systems that use "soft" control laws with smooth trajectory generation, the robot can use the natural dynamics of its mechanical structure to move through a gait pattern. The work in this paper shows sound walking performance that can only improve with the augmentation of global inertial sensors and feed-back paths. References [McMillan, 1995] S. McMillan, Computational Dynamics for Robotic Systems on Land and Underwater, PhD Thesis, Ohio State University, 1995. [Nishiwaki, 2000] K. Nishiwaki, T. Sugihara, S. Kagami, F. Kanehiro, M. Inaba and H. Inoue, Design and development of research platform for perception-action integration in humanoid robot: H6, International Conference on Intelligent Robots and Systems, IROS 2000 [Wyeth, 2001] G. Wyeth, D. Kee, M. Wagstaff, N. Brewer, J. Stirzaker, T. Cartwright, B. Bebel. Design of an Autonomous Humanoid Robot, Proceedings of the Australian Confer- ence on Robotics and Automation (ACRA 2001), 14-15 November 2001,Sydney [Yamaguchi, 1998] J. Yamaguchi, S. Inoue, D. Nishino and A. Takanishi, Development of a bipedal humanoid robot having antagonistic driven joints and three DOF trunk, Proceedings International Conference on Intelligent Robots and Systems, IROS 1998 [Yamasaki, 2000] F. Yamasaki, T. Matsui, T. Miyashita, and H. Kitano. PINO the Humanoid that Walk, Proceedings of First IEEE-RAS International Conf on Humanoid Robots, CDROM 2000 附录2 分布式控制的步态的人形机器人。 戈登惠氏和达米安记 学校的信息技术和电气工程 昆士兰大学，圣卢西亚，昆士兰州， 4072 ，澳大利亚 (惠氏，达明安) @ itee.uq.edu.au 摘要。本文介绍一种步态的人形机器人分布式控制系统。该项议案的机器人实时计算的 中央控制器，并派遣了CAN总线的分布式控制系统。那个分布式控制系统松散如下议案模式 从中央控制器，同时也保持稳定和平衡。不存在全球反馈控制系统;该系统保持平衡的互动中 欧之间的步态和"软"控制的驱动器。该文件说明直线行走步态与 表 关于同志近三年现实表现材料材料类招标技术评分表图表与交易pdf视力表打印pdf用图表说话 pdf 演之间的相互作用步态代 和控制系统。数据的分析表明，成功的走路也可以达到严格地方联合控制没有明确的全球平 衡协调。 1简介 人形机器人协调控制通常需要大量的关节。印第安纳州现有的大多数实现人形机器人，协调 所取得的使用一个中央控制计算机的接口，所有的传感器和驱动器提供局部控制的共同立场 和扭矩以及全球的平衡与控制态势。本文介绍了一种分布式的方法来控制和协调， 提供本地 控制的立场和扭矩在每个联合的方式，保持全球平衡和姿态。 1.1纸概况 在简要介绍了有关的工作，介绍了GuRoo机器人形式的基础，后来的实验。细节的建筑和 设计 领导形象设计圆作业设计ao工艺污水处理厂设计附属工程施工组织设计清扫机器人结构设计 的机器人，然后有一段描述运算系统，支持分布式控制系统。然后，本文件描述的方法，分布式控制和提供详细的步态一代，包括结果收集到一条直线步行。 2相关工作 该OpenPino项目[山崎拓， 2000 ] ，采用了非常集中的方式，以人形控制。板载SH2结构微控制器负责采取高命令从PC ，如步行前进，停止等，并将其转换为位置4 Polani等信息。这些立场是转换成PWM信号由一个单一CLPD ， 负责控制所有26个自由度。 早稻田大学的人， WABIAN ，采用了类似的控制系统， 与高层次的命令所产生的板载奔腾166Mhz计算机[ 山口县， 1998 ] 。由此产生的速度剖面，被输入两个16 通道数/董事会通过ISA总线，而供应的电机驱动器所需的信号以驱动汽车。这个系统是身体上的集中与所有的计算设备和电机驱动器位于躯干。 同样， H6由东京大学利用板载PIII 700Mhz 计算机产生高水平的命令[西， 2000 ] 。一对富士通的I / O 控制器，类似WABIAN的，产生控制信号所必需的汽车。 单独电机驱动器提供必要的权力位于身体接近发动机。 2.1 GuRoo项目 GuRoo是一个1.2米高，人形机器人完全自主设计并建造的昆士兰大学机器人实验室[惠氏， 2001 ] 。机器人总大众34公斤，其中包括机上电源和计算。目前能够GuRoo 一些示威任务包括平衡，散步，车削，卧虎藏龙， 握手和挥手。该机器人已进行现场演示组合 这些任务在不同的机器人表演。 打算挑战任务机器人玩游戏或足球对人类的球员或其他人机器人。要完成这一挑战， 机器人必须能够自由流动，其两条腿。显然，机器人必须运作 完全自主的方式不支持或电线线束系绳。那个目前GuRoo机器人无法承受的影响与踢足球，但充当了一个很好的平台，研究设计的平衡行为动态步态控制的位置和轴驱动每一联合中可以看出 图1 图。 1 。人形机器人的GuRoo和自由度每一联合。 2.2机电设计 关键因素推动了机械设计的选择器。 该机器人有23个关节的总额。腿部和腹部包含有15个关节分布式控制的步态的人形机器人705须出示重大机械动力，最普遍的大扭矩和相对较低的速度。其他8关节驱动器的头部和颈部大会，并武器与大大低于扭矩和速度的要求。 高功率的15个关节都使用相同的发动机变速箱的组合构成一个36马克森稀土电机与变速箱减少了156:1 。最大持续产生的输出扭矩是10纳米。每个电机装有光学编码位置和速度反馈。低功耗的8关节高新技术钢筋混凝土伺服电机模型HS705球蛋白。额定输出扭矩140牛米。 马达驱动辊轴的髋关节还辅以弹簧与弹簧常数1海里/度。这些泉水有助于抵消自然趋势腿相撞，并帮助产生晃动动议是关键成功的步态。 电源提供了2 × 1.5Ah镍镉包芯片则内置42V的高功率电机， 2 x 3Ah 7.2 V镍镉电池的计算和伺服操作。该包 选择给予20分钟的连续运转。 2.3遥感 位置反馈编码器的高功率接头提供867编码器每程度的联合动议。此外，每个DSP可以测量电流每个电机。还编列经费和平衡的惯性传感器，作为以及联络开关的脚和关节。 3分布式控制网络 一个分布式控制网络控制机器人，与一个中央计算机中心， 设定目标的机器人，传感器信息处理，并提供协调目标关节。关节都有自己的控制处理器，这种行为在团体保持全球稳定， 同时也经营单独提供本地马达控制。分布式系统是连接的CAN网络。 3.1中央控制 的中央控制机器人推导了联合速度须完成步态。阿PIII 1.1GHz的笔记本电脑目前的计算速度为所有23度自由实时。的速度沿传递串行链接到一个分配董事会作为之间的桥梁串行总线与控制区域网络(加拿大)已提供经费，以港口为这种控制安装在康柏的iPaq 机器人，以实现真正的自治，没有任何系绳。 3.2联合控制器 所有联合控制器执行使用TMS320F243数字信号处理器德州仪器，一个16位DSP设计的电机控制。提供的CAN模块在这一系列，同时引导可编程内部闪存记忆使该器件特别有吸引力的这一申请。五年控制器议会控制15个高功率马达，每局3马达控制。第六控制板控制8钢筋混凝土伺服电机。图2概括了互动各节点之间的控制网络。 CAN总线 串行 图。 2 。框图分布式控制系统。 4软件 该软件包括四个主要实体:全球运动一代代码， 当地的电机控制，在低级别的代码的机器人，以及模拟器。该软件组织提供了一个标准的接口都是低级别的代码，并在机器人 模拟器。这意味着，在仿真软件开发可简单编译经营的实际机器人。因此，该机器人采用了一些标准的接口要求是用于机器人和模拟器包括阅读的编码器和设置的PWM值以及转让的CAN数据包。 4.1步态生成 步态生成模块负责产生变现的轨迹23个关节，使机器人可以执行基本的行为，如站立在一条腿， 卧虎藏龙和散步。最重要的属性是，它们的轨迹正在顺利进行，他们可以联系在一起平稳时尚。光滑性议案意味着较少干扰控制其他关节。根据这些标准， 归一化联合调度如图3所示，适用于所有机动轨迹..的运动轨迹是一个 参数 转速和进给参数表a氧化沟运行参数高温蒸汽处理医疗废物pid参数自整定算法口腔医院集中消毒供应 化产生正弦曲线在哪里理想的关节速度，是全关节角度移动和T是这一时期的运动。的轨迹，与典型的机器人轨迹生成 机械手，通常集中在平滑的运动而不是结束效应比光滑的联合。 轨迹为每个电机可协调使用相同的开始时间的议案，并指明同一时期的轨迹。轨迹可能是自然联系的速度都达到零的开始和年底的议案。第5节将说明如何轨迹可能是协调和联系执行行走作业。 分布式控制的步态的人形机器人707 4.2联合控制器软件 有两种类型的联合控制器板中使用的机器人摄氏度五个控制器 议会控制15高功率马达和一个控制器控制的8个低 功率电动机。控制器软件的低功率电动机是一个单一的中断 这是例行的到来所引发的CAN数据包给控制器,问 信箱。常规内容的CAN信箱改变立场发出的 步态代常规。的PWM占空比控制的立场，钢筋混凝土 舵机是多种多样的相应。控制回路的高功 率控制器有两个中断例程。至于低功耗控制器，中断执行，在收到的轨道数据 CAN总线邮箱。这些数据是用来设置速度设定点的电机控制例行公事。此外，还定期中断每500 s的运行电机控制软件。 常规的电机控制比较误差和速度设定编码器阅读和生成的PWM值为电机的比例， 积分控制法。日常检查也电机电流对电流限制和调整的PWM值，以防止过电流的情况。的PI控制法，每一联合已另一方面调整，以提供良好的轨迹以下为典型的速度输入个人资料，和弹簧阻尼器模型阻抗以扭矩干扰严重性和交叉偶联其他关节力矩。 "软"回应这种控制法，干扰阻碍扭矩传送整个机器人，并有助于维护全球稳定。缺点是该控制器患有位置误差，必须占的步态代软件。第5节说明了这个潜在的赔偿责任是转向资产在新一代的动态稳定行走。 4.3低阶程式码 最低级别的代码机器人提供直接接触的传感器和通信系统。抽象的水平所提供的函数调用在这个级别艾滋病的交叉发展的代码之间的模拟器和真正的机器人。 4.4模拟器 该模拟器是基于DynaMechs项目[麦克米兰， 1995 ] ，与补充模拟特定功能的机器人，如直流电动机和驱动马达的钢筋混凝土舵机，该传感器，异构处理环境与CAN网络。 这些增加提供相同的接口，动态图形仿真作为联合控制器和步态生成代码。的参数模拟器来自CAD模型和数据表从已知组成部分。这些参数包括修改Denavit - Hartenberg参数这说明机器人的拓扑结构，矩阵的张量的联系和各种电机和变速箱的特点与每个联合。从表面数据的CAD 模型也是进口的模拟器的图形显示。对于高功率直流电动机关节，模拟器提供了编程读数从编码器及电流传感器的基础上，速度和扭矩从动力学方程。对于钢筋混凝土舵机，模拟器更新的立场的基础上关节PD模型与有限的转换率。那个程序员必须提供仿真的PWM值为汽车提供控制。模拟器提供虚假中断，以模拟真实事件的根据控制软件。 该模拟器采用了一体化的步长为500 s和更新的图形显示每5毫秒的模拟时间。当运行在1.5 GHz的奔腾4在Windows 2000年，模拟更新所有23关节在一个非常有用的40 ,的实时速度。 5散步 nuvo可以行走一步步率为1赫兹使用步长100毫米。步行是开环;没有反馈联合控制器的步态代软件。缺乏全球性的反馈，加上没有一个全球性的平衡传感器是一个重大的挑战在步行算法设计。 5.1散步算法 该机器人采用了简化版本的一个典型人类的步态。特别是，它限制了摆动的腿，以防止平衡紊乱，因为这不能被纠正全球平衡控制。为了尽量减少加速度的躯干，头武器(其中占1 / 3的质量机器人) ，机器人保持不变相对位置的躯干，因此面对躯干始终是正常的方向。允许的唱名的躯干也很有限。稳定躯干也减少干扰重力的控制下腿关节。在散步，每个机器人的运动负荷对重力表演略有蹲是引进了6度踝间距，与膝关节和髋关节间距设置为保持身体直立。最初载入中关节减少反弹的可能性在gearheads 。行走步态开始与侧侧摆产生的辊轴的脚踝和臀部。的摆动频率为0.5赫兹是同情的弹簧质量系统所形成的踝关节控制器与大众的机器人。听命成立格局的重量转移到一英尺的其他必要的回旋腿交替实现散步。在每一个极端的摇摆，惯性上机构确保ZMP点(零力矩点)的机器人在于支持多边形 所形成的支持脚，甚至认为该中心的质量可能不会。这项行动地方需要较少的扭矩从髋关节和踝关节辊驱动器，因为该动议由于重力使机器人远离极端每个摇摆。 图。 4 。正面观的行走过程。 一旦动摇足以留下任何地面反作用力的非支持脚，非支撑腿取消使用间距轴的髋，膝和踝关节。 之间的升降非支撑脚，每个偏航轴电机曲折，因此，非支撑腿摆动着创造的一步。当 摆动腿接触地面，机器人是动态的稳定，质量中心在支持脚在额面，但在前面的脚趾在矢状 飞机。该机器人然后波动影响到其他脚，重复序列和进展与步行.. 5.2分析结果 该项议案的机器人是最好的分析比较理想的速度从步态一代模块的实际速度在每个联合。图5显示这比较该议案的髋，膝和踝关节的唱名，音高和偏航轴。那个图形初始化中段双拥阶段，双腿在与地面接触。图表包括一秒的数据，描述的权利腿部动作，因为它从双拥阶段，通过摆动期回到双拥阶段。点吨= 0.25 s时，摆动腿开始解除和失去与地面接触。 与髋辊轴的摆动腿不再有助于支持机器人，在这个春天位于简要主导轴的实际速度造成过阻尼振荡见于髋关节在这个时候。 一旦脚离开地面的脚踝辊电机开关驾驶的腿从脚，推动徒步腿。这大幅度减少，相对惯性结果在一份简短的增加，规模的脚踝辊速度。脚下有一个相对较低的惯性，而其余的机器人，这样的PI控制器没有什么麻烦下列所需的速度，直到脚再次使接触地面。该机器人到达极端每个晃动在t = 0.5秒，在所有运动在飞机停止滚动。摆动腿现在理论上完全解除，尽管膝关节和髋关节间距没有达到他们想要的职位，直到? = 0.6s 。 图。 5 。理想与实际联合角度直线行走超过一秒时期。图 期间开始的双重支持和后续阶段右腿通过摇摆阶段。实际联合流速剖面为每个摊位电机的摆动腿显示了越来越多影响的严重性和腿的摆动惯性阶段。积分任期的PI控制器旨在消除稳态误差，因此，占主导地位的实际速度，每个摊位电动机驱动的理想位置。随着马达不 达到的最高理想的速度，它是被迫延长运动时间确保每一个地区的图都是平等的。长期的低比例的结果是穷人跟踪期望的速度，而且使联合，以更好地应对外部骚乱。 比较实际的速度与预期的速度为每一个球运动在每个回合降低从脚踝到膝盖的髋关节。踝关节只需要 加快脚，而必须加速膝关节的脚和小腿。髋关节球场必须加快整个腿在摆动期。该项议案的偏航轴的摆动腿的取消，推动机器人前进。 当发生偏航运动的支持腿的势头机器人原因 联合过度其立场。摆动腿下，然后把机器人成双重支持的阶段。摩擦的2英尺对地面和重量机器人腿的支持防止偏航轴定位误差得到解决。 这提供了一个预先载入的联合，支持这项动议的未来的偏航挥杆阶段。机器人左右回到另一边，体重逐渐释放支持脚，直到扭矩代理的联合克服了共同有效的脚之间的摩擦和发言。这不一定是在该点摆动腿失去与地面接触。时间的腿已经解决了这个错误， 联合正在经历偏航运动与摆动腿扭曲。因此，该曲线下面积为髋偏航在摆动期大于理想的地区。 与地面接触，实现在T = 0.85s ，一旦作出，机器人返回双重支持阶段的步态。无论是髋关节和踝关节的摆动腿现在协助支持腿轧辊电机动摇整个机器人的其他脚，并在进程逐步开关的作用支持和摆动腿 6结论 本文表明人形机器人可以行走，而不需要明确全球的信息反馈，或严格控制的共同轨迹。通过结合一组松散的协调控制系统，使用"软"控制法与平滑轨迹代，机器人可以利用其自然动态的机械结构，以动议通过步态模式。这项工作文件表明健全步行性能只能改善，增加全球惯性传感器和反馈路径。 参考资料 [麦克米兰， 1995 ]美国麦克米兰，计算动力学机器人系统的土地和水下，博士论文，美国俄亥俄州立大学， 1995年。 [西， 2000 ]光西，吨杉原，南镜，楼兼先生稻和H.井上， 设计与开发的研究平台，知觉行动一体化人形机器人: H6 ，国际会议的智能机器人与系统， IROS 2000年 [惠氏， 2001 ]湾惠氏， 4记先生瓦格斯塔夫，北布鲁尔，学者Stirzaker ，吨卡特赖特湾倍倍尔。设计自主仿人机器人，议事会议的澳大利亚机器人与自动化( ACRA 2001 ) ， 14-15 2001年11月，悉尼 [山口县， 1998年]学者山口，由井上， 4西和A. Takanishi ，发展双足人形机器人有拮抗驱动三自由度关节和躯干，议事国际会议的智能机器人与系统， IROS 1998年 [山崎拓， 2000 ]楼山崎拓，松井，吨宫，和H.北野。皮诺阿的人走路，第一次录的IEEE - ras基因国际配置的人型机器人， 光盘2000年 徐州工程学院毕业设计(论文) 徐州工程学院毕业设计(论文) 52 43