📚 核心词汇记忆链
7章逻辑记忆链 · 57个核心词汇 · 更新日期:2026-04-16
🤖 第一章 基础概念(7词)
【中文】【中文】机器人 — 能自动执行复杂动作序列的可编程机器
EN: A machine capable of carrying out complex series of actions automatically, especially one programmable by a computer; senses, thinks, and acts in a coordinated loop.
🔗 🔗 近义:automaton(自动机,较老派);android(人形机器人);cyborg(半机械人)
📝 📝 The modern robot can perform assembly tasks with sub-millimeter precision, surpassing human capability in repeatability.
→ → 现代机器人能够以亚毫米精度执行装配任务,在重复精度上超越人类。
💡 💡 源自捷克语 robota(强制劳动)。1920年卡雷尔·恰佩克在《罗莎的万能机器人》(Rossum's Universal Robots) 中首次使用。1954年德沃尔申请了全球第一份机器人专利,1961年Unimate首次安装在通用汽车压铸车间。
【中文】【中文】机器人学 — 设计、构造、操作和应用机器人的科学技术领域
EN: The interdisciplinary branch of engineering and science that deals with the design, construction, operation, and use of robots — encompassing mechanical engineering, electrical engineering, computer science, and cognitive science.
📝 📝 Robotics has revolutionized automotive manufacturing, reducing production costs by up to 50% while improving weld quality consistency.
→ → 机器人学已经彻底改变了汽车制造业,在提升焊缝质量一致性的同时将生产成本降低了高达50%。
💡 💡 robotics = robot + -ics(学科后缀)。注意:永远作单数使用(不说 "a robotics")。学科三大支柱:运动学(几何)、动力学(力)、控制理论(算法)。
【中文】【中文】自动化 — 用自动设备替代人工劳动、提高产出一致性的技术
EN: The use of automatic equipment and control technologies (not necessarily robots) to reduce or eliminate human labor in production, logistics, or information processing.
🔗 🔗 关联:industrial automation(工业自动化);office automation (OA)(办公自动化);process automation(流程自动化)
📝 📝 Factory automation has dramatically lowered production costs in developed economies, but has also displaced millions of manufacturing jobs.
→ → 工厂自动化大幅降低了发达经济体的生产成本,但也取代了数百万制造业岗位。
💡 💡 auto(自己)+ -ation(行为结果)。机器人是自动化的子集,但自动化≠机器人。PLC控制的传送带、自动化仓储系统、自动售货机都是自动化但不是机器人。
【中文】【中文】机械臂 — 机器人中负责在空间中执行物理操作的手臂机构
EN: The articulated arm mechanism of a robot — typically comprising a base, shoulder, elbow, wrist, and end effector — that physically performs tasks by moving in 3D space.
🔗 🔗 全称:robot arm(机械臂);robotic arm-manipulator system(机械臂操纵系统)
📝 📝 A standard 6-axis manipulator can reach any point and orientation within its spherical work envelope, enabling complex tasks like arc welding in 3D space.
→ → 标准6轴机械臂可以到达其球形工作空间内的任意位置和姿态,使三维空间弧焊等复杂任务成为可能。
💡 💡 来自拉丁语 manipulare(用手操作)。机械臂特指手臂本体,不含底座(通常固定)和末端执行器(工具)。类比:人手臂=manipulator,手掌=end effector,手指=具体的执行工具。
【中文】【中文】末端执行器 — 安装在机械臂末端的工具,机器人直接与外界交互的接口
EN: The tool, device, or mechanism attached to the wrist of a robot arm that physically interacts with the work environment — grippers, welding torches, paint sprayers, sanders, magnetic lifters, etc.
🔗 🔗 同义:tool(工具,非正式);EOAT(End of Arm Tooling);wrist attachment
📝 📝 The same 6-axis robot can perform pick-and-place, welding, and screw-driving simply by swapping the end effector — a gripper for picking, a welding torch for joining, a screwdriver for fastening.
→ → 同一台6轴机器人只需更换末端执行器就能执行抓取、焊接和螺丝拧紧——夹爪负责抓取,焊枪负责连接,螺丝刀负责紧固。
💡 💡 机器人的"手"。同一机械臂换不同末端执行器=执行完全不同任务。选型关键:负载能力(夹爪自重占负载)、精度要求(焊枪精度要求极高)、换装时间(快换接口决定柔性)。
【中文】【中文】自由度 — 机器人机构能独立运动的方向数量(每个关节=1 DOF)
EN: The number of independent parameters (translational or rotational) that define the configuration of a mechanical system; in robotics, each joint represents 1 DOF. A rigid body in 3D space has 6 DOF (3 translation + 3 rotation).
🔗 🔗 简称:DOF;redundant DOF(冗余自由度,人手臂7 DOF);underactuated(欠驱动,DOF不足)
📝 📝 A human arm has 7 DOF (shoulder 3 + elbow 1 + wrist 3), while a standard 6-axis industrial robot has 6 DOF — enough to reach any position/orientation but with no redundancy.
→ → 人手臂有7个自由度(肩3+肘1+腕3),而标准6轴工业机器人恰好有6个自由度——足以到达任意位置/姿态,但没有冗余。
💡 💡 3个移动(XYZ) + 3个旋转(俯仰/偏航/翻滚) = 6 DOF = 空间刚体的最小完整配置。人手臂7 DOF多出的1个让肩部运动有解的多样性(即冗余性)。少一个维度→无法任意姿态到达目标点。
【中文】【中文】工作空间 — 机器人末端执行器所能到达的所有空间点的集合
EN: The complete set of positions that the robot's end effector can reach, determined by the arm geometry, joint limits, and physical obstacles. Also called reach space or work volume.
🔗 🔗 同义:work volume(工作体积);reach space(可达空间);dexterous workspace(灵巧工作空间,指所有可达且可任意姿态到达的区域)
📝 📝 A robot with a large work envelope is essential for tasks like aircraft wing painting — the entire wing surface must be reachable without repositioning the aircraft.
→ → 工作空间大的机器人对于飞机机翼喷漆等任务至关重要——必须能在不重新定位飞机的情况下触达整个机翼表面。
💡 💡 选机器人第一步:工作空间能不能覆盖任务点。SCARA=圆柱形工作空间,关节型=球形,并联Delta=倒金字塔。记住:工作空间大≠好,小而精确有时更实用。
⚙️ 第二章 核心零部件(14词)
【中文】【中文】执行器 — 将电/液压/气压能量转换为机械运动的装置,机器人的"肌肉"
EN: A transducer that converts energy (electrical, hydraulic, or pneumatic) into controlled mechanical motion — the prime mover of a robot joint.
🔗 🔗 三大类型:electric actuator(电动,伺服电机);hydraulic actuator(液压,大力矩);pneumatic actuator(气动,轻载高速)
📝 📝 Servo motors are the most common actuators in precision robotics because they combine high torque density with precise position control — critical for smooth, accurate motion.
→ → 伺服电机是精密机器人中最常见的执行器,因为它们将高扭矩密度与精确位置控制结合——这对于平滑精确的运动至关重要。
💡 💡 机器人的"肌肉"。选错执行器类型直接决定应用场景:重载液压(挖掘机/飞机模拟器)、轻载高速气动(食品分拣)、精密可控电动伺服(几乎所有现代工业机器人+人形机器人)。
【中文】【中文】伺服电机 — 带闭环位置反馈的高精度电动执行器,"想停哪就停哪"
EN: A motor specifically designed for precise control of angular or linear position, velocity, and acceleration — using a built-in encoder for closed-loop feedback to a controller (PID or similar).
🔗 🔗 全称:servomechanism motor;简称:servo;对比:stepper motor(步进电机,开环控制,便宜但精度低)
📝 📝 The robot's joints are all driven by high-torque servo motors with integrated 17-bit absolute encoders, enabling sub-arc-minute positioning accuracy.
→ → 机器人的关节全部由带17位绝对值编码器的集成高扭矩伺服电机驱动,实现亚弧分定位精度。
💡 💡 "伺服"=servomechanism(伺服机构)。关键特征:闭环控制+编码器位置反馈。普通电机通电就转,伺服电机受控"想停哪就停哪"。三大关键参数:扭矩(Nm)、转速(RPM)、惯量匹配(决定动态响应)。
【中文】【中文】编码器 — 将机械位置转换为数字/模拟信号的传感器,机器人的"本体感觉"
EN: An electro-mechanical sensor that converts angular or linear position into electrical signals (digital pulses or absolute code) for the controller to determine exact joint angle — the robot's proprioception.
🔗 🔗 类型:incremental encoder(增量式,只报相对变化);absolute encoder(绝对值,断电记忆位置);resolver(旋变,模拟式抗干扰强)
📝 📝 An absolute encoder retains the exact joint angle in memory — even after power loss — so the robot does not need to perform a homing reference run when restarted.
→ → 绝对值编码器能在断电后依然保留精确的关节角度数据——因此机器人重启时不需要执行回原点操作。
💡 💡 想象成机器人的"本体感觉"——人闭眼也知道手在哪,机器人靠编码器知道关节转到了哪里。增量式每次开机需要"回零",绝对值编码器断电不丢位置。分辨率用"位"表示:17位=2^17=131072分辨率/圈。
【中文】【中文】减速器 — 降低转速、增加扭矩的机械传动装置,机器人关节的"力量放大器"
EN: A mechanical transmission device that reduces motor speed (RPM) while proportionally increasing output torque — essential because robot motors spin too fast (1000-3000 RPM) for direct drive, and need torque amplification.
🔗 🔗 两大类:RV reducer(摆线针轮,承载力强,1-3轴用);harmonic reducer(谐波,轻量精密,4-6轴用);对比:planetary gearbox(行星减速器,成本低)
📝 📝 The RV reducer is typically used for robot axes 1-3 (base, shoulder, elbow) because of its high torque capacity and rigidity, while the lighter harmonic reducer handles wrist axes 4-6.
→ → RV减速器因其高扭矩承载和刚性通常用于机器人1-3轴(基座、肩部、肘部),而较轻的谐波减速器负责腕部4-6轴。
💡 💡 机器人电机转速1000-3000 RPM,直接驱动太猛太快。减速器=降速增扭。谐波(轻载精密,零背隙)+RV(重载稳健)是工业机器人两大核心。并联机器人(Delta)一般不用RV,用行星或皮带传动。
【中文】【中文】谐波减速器 — 基于柔性轮弹性波变形原理的精密减速器,"零背隙"是其核心优势
EN: A precision strain wave gearing system using controlled elastic deformation of a flexible (wave) generator to achieve high reduction ratios with zero backlash — characterized by compact size, light weight, and high precision.
🔗 🔗 别名:harmonic gear;结构三件套:wave generator / flexspline / circular spline;主要供应商:哈默纳科(日)、绿的谐波(中)
📝 📝 Harmonic drives are preferred for wrist axes (4-6) due to their zero backlash (no gear clearance causing positioning error) and compact size — critical for the precision required in assembly tasks.
→ → 谐波减速器因其零背隙(无齿轮间隙导致的定位误差)和紧凑结构常用于腕部关节——这对于装配任务所需的精度至关重要。
💡 💡 三大供应商:哈默纳科(日本垄断高端)、绿的谐波(中国挑战者)、来福谐波(中国)。原理独特:不是齿轮咬合,是"柔轮弹性波"变形传递动力,区别于所有传统减速器。人形机器人关节=谐波减速器刚需(轻+紧凑+零背隙)。
【中文】【中文】机器人控制器 — 机器人的大脑,实时执行运动程序、处理传感器数据、命令所有执行器
EN: The industrial PC-based or embedded computer that executes robot motion programs (trajectory planning, interpolation), processes sensor feedback, and commands all joint actuators in real time via high-speed communication (EtherCAT, PROFINET).
🔗 🔗 代表产品:发那科FANUC R-30iB、ABB OmniCore、库卡KRC5;编程语言:KAREL(发那科)、RAPID(ABB)、KRL(库卡)
📝 📝 The robot controller executes the taught path by interpolating thousands of intermediate points per second, ensuring smooth motion even when the operator only taught key waypoints.
→ → 机器人控制器通过每秒插补数千个中间点来执行示教轨迹,确保即使操作员只示教了关键路径点,机器人也能平滑运动。
💡 💡 机器人的大脑。性能决定轨迹精度和响应速度。发那科用自研FANUC控制器;ABB用OmniCore(全新平台);国产机器人多用EtherCAT总线+工业PC(低成本但生态弱)。控制器=硬件+运动控制算法+通信协议,三者缺一不可。
【中文】【中文】传感器 — 检测环境变化(位置/力/视觉/距离)并向控制器发送信号的装置,机器人的"五感"
EN: A device that detects physical quantities (position, force/torque, proximity, vision, distance, temperature, vibration) and converts them into electrical signals for the robot controller to make decisions.
🔗 🔗 分类:proprioceptive(本体感知,内部:编码器、IMU);exteroceptive(外部感知,外部:相机、力矩、超声);force-torque sensor(力矩传感器);proximity sensor(接近传感器)
📝 📝 A 6-axis force-torque sensor in the robot wrist enables peg-in-hole assembly — the robot detects contact forces and adjusts alignment in real time, compensating for millimeter-level positional errors.
→ → 机器人腕部的六轴力矩传感器使"插销入孔"装配成为可能——机器人实时检测接触力并调整对准,补偿毫米级的位置误差。
💡 💡 机器人五感:视觉(工业相机)、力觉(力矩传感器)、触觉(电子皮肤)、听觉(麦克风阵列)、嗅觉(气体传感器)。没有传感器,机器人就是瞎子聋子——只能按固定程序走,无法应对任何变化。
【中文】【中文】无刷直流电机 — 无碳刷机械接触的直流电机,高效率、高功率密度,人形机器人关节驱动主流
EN: A DC motor that uses permanent magnets and electronic commutation (via ESC or controller) instead of brushes — offering high efficiency, high power density, low heat, and long life at the cost of more complex control electronics.
🔗 🔗 别名:brushless DC / BLDC;对比:brushed DC(有刷,便宜但效率低);近义:PMSM(永磁同步电机,控制方式更复杂但更平滑)
📝 📝 Tesla Optimus uses custom BLDC motors in each joint — avoiding gear backlash by using low-reduction, high-torque direct drive enabled by the high power density of modern BLDC technology.
→ → 特斯拉Optimus每个关节使用定制无刷直流电机——利用现代无刷电机的高功率密度实现低减速比、高扭矩直驱,避免齿轮背隙问题。
💡 💡 有刷电机碳刷会磨损、有火花、效率低。无刷=没有机械接触=寿命长、效率高、转速范围宽。人形机器人关节因为需要高功率密度+低惯性+高响应,几乎全部使用无刷电机+低减速比谐波减速器的组合。
【中文】【中文】步进电机 — 将电脉冲精确转换为离散转角的执行器,开环位置控制
EN: A brushless DC motor that divides one rotation into a large number of discrete steps — each current pulse advances the shaft by one step angle (e.g., 1.8°), enabling open-loop position control without encoder feedback.
🔗 🔗 类型:PM stepper(永磁步进);VR stepper(可变磁阻步进);hybrid stepper(混合式,精度最高);对比:servo(闭环,更高精度)
📝 📝 Stepper motors dominate 3D printers and CNC routers because they offer good accuracy at low cost in open-loop — no encoder needed, simpler controller, lower system cost.
→ → 步进电机在3D打印和CNC雕刻机中占主导地位,因为它们在开环控制下以低成本提供良好精度——无需编码器,控制器更简单,系统成本更低。
💡 💡 步进=脉冲数=精确转角。开环控制(不需编码器反馈)=成本低、系统简单。但丢失脉冲不自知(负载过大时"丢步"),不适合作业型机器人关节。精密机器人用伺服,定位机构(3D打印、雕刻)用步进。
【中文】【中文】惯性测量单元 — 测量物体三轴角速度和加速度的装置,机器人的"前庭系统"
EN: An electronic device that measures linear acceleration and angular rate (and often magnetic field) using accelerometers, gyroscopes, and magnetometers — used for attitude estimation, balance control, and dead reckoning in robots.
🔗 🔗 组成:3-axis accelerometer(加速度计)+ 3-axis gyroscope(陀螺仪);融合算法:AHRS(姿态航向参考系统);IMU vs VRU vs AHRS区别在于是否含磁力计
📝 📝 Boston Dynamics Atlas uses an IMU in its torso combined with leg kinematic data to maintain balance during dynamic maneuvers like jumping and backflips — the IMU provides millisecond-level body attitude feedback.
→ → 波士顿动力Atlas在躯干中使用IMU,结合腿部运动学数据维持跳跃和后空翻等动态机动中的平衡——IMU提供毫秒级躯干姿态反馈。
💡 💡 类比人的前庭系统(内耳),感知头部的倾斜和旋转。双足机器人站立和行走时,IMU反馈躯干姿态,控制器实时调节关节扭矩维持平衡(ZMP/WBC控制)。没有IMU,双足机器人无法在不平地面行走。
【中文】【中文】力矩传感器 — 测量机器人腕部三维力和三维力矩的传感器,"触觉"的精密形式
EN: A precision transducer (typically strain-gauge based) installed at the robot wrist that measures 3 forces (Fx, Fy, Fz) and 3 torques (Tx, Ty, Tz) — enabling force-controlled assembly, deburring, polishing, and collaborative safety.
🔗 🔗 别名:6-axis F/T sensor(六轴力矩传感器);著名品牌:ATI(美国)、JR3(美国)、Omega(德国ATENSOR);安装在:wrist flange(腕部法兰)
📝 📝 A robot equipped with a force-torque sensor can perform precision assembly like "peg-into-hole" — the robot monitors contact forces in real time and adjusts insertion depth to avoid wedging, eliminating the need for sub-millimeter positioning accuracy.
→ → 配备力矩传感器的机器人能够执行"插销入孔"等精密装配——机器人实时监控接触力并调整插入深度以避免卡滞,无需亚毫米级定位精度。
💡 💡 没有力矩传感器的装配=盲插。力矩传感器让机器人感知"碰到东西了/太紧了/歪了"。应用:精密装配(精密齿轮插入)、表面加工(打磨去毛刺)、协作机器人碰撞检测(人碰到时立即停止)。
【中文】【中文】接近传感器 — 检测物体是否进入指定范围内的非接触式传感器
EN: A non-contact sensor that detects the presence or absence of objects within a specified range using magnetic, inductive, capacitive, or optical principles — widely used for object detection, position verification, and collision avoidance.
🔗 🔗 类型:inductive(电感式,金属检测);capacitive(电容式,非金属检测);photoelectric(光电式);ultrasonic(超声波)
📝 📝 An inductive proximity sensor mounted near the robot's gripper detects whether a metal workpiece is correctly seated before the gripper releases it — a simple binary check that prevents costly errors.
→ → 安装在机器人夹爪附近的电感式接近传感器检测金属工件是否正确就位,然后再由夹爪释放——一个简单的二元检查,防止代价高昂的错误。
💡 💡 工业现场最常见的传感器之一。作用:到位检测(气缸是否推出)、存在性检测(零件是否在正确位置)、限位检测(代替机械限位开关)。选型:金属工件→电感式;非金属→光电或电容式。
【中文】【中文】夹爪 — 机器人的抓取末端执行器,通过机械爪或吸盘拾取和释放物体
EN: The end effector specifically designed for grasping and releasing objects — including 2-finger parallel grippers, 3-finger centric grippers, angular grippers, and vacuum grippers (suction cups) for smooth or non-magnetic surfaces.
🔗 🔗 类型:parallel jaw gripper(平爪);angular gripper(角度爪);vacuum gripper(真空吸盘);soft gripper(软体夹爪,仿生);magnetic gripper(磁性夹爪)
📝 📝 A soft pneumatic gripper with silicone fingers can handle delicate fruits like strawberries without bruising — impossible with rigid metal fingers that apply excessive point pressure.
→ → 带有硅胶手指的柔性气动夹爪能够轻柔处理草莓等易损水果而不造成压伤——这是刚性金属手指无法做到的,因为会产生过大的点压力。
💡 💡 最常见的末端执行器。选型关键:抓取力(夹持力N)、行程(爪开合范围mm)、定位精度、适用工件形状(圆形/方形/不规则)、表面材质(光滑→真空吸盘)。柔性夹爪是食品/医药行业的刚需。
【中文】【中文】工业总线 — 机器人控制器与关节驱动器之间的实时通信网络
EN: Real-time industrial Ethernet protocols (EtherCAT, PROFINET, EtherNet/IP) used to synchronize joint controllers, sensors, and actuators with microsecond-level latency — the "nervous system" of modern factory robots.
🔗 🔗 主流协议:EtherCAT(倍福,Beckhoff,最流行);PROFINET(西门子);EtherNet/IP(罗克韦尔);CANopen(较早,低成本);对比:OPC-UA(上层信息交互,非实时)
📝 📝 EtherCAT achieves 100 μs cycle time with less than 1 μs jitter, enabling synchronized control of 16+ robot joints — critical for coordinated motion like synchronizing dual-arm assembly tasks.
→ → EtherCAT实现100微秒周期且抖动小于1微秒,支持16个以上机器人关节的同步控制——这对于双臂协调装配等同步运动至关重要。
💡 💡 实时性=一切。机器人控制需要毫秒甚至微秒级响应,工业总线=保证所有关节在同一时钟下同步运动。EtherCAT是目前工业机器人最流行的协议,循环时间可达100微秒(1/10000秒)。
🎯 第三章 运动控制(7词)
【中文】【中文】运动学 — 研究运动(位置/速度/加速度)本身,不考虑产生运动的力
EN: The branch of mechanics describing the motion of objects without considering the forces that cause the motion — in robotics, describes how joint angles translate into end-effector position/orientation.
🔗 🔗 对比:dynamics(动力学,研究力与运动的关系);forward kinematics(正运动学,已知关节角→末端位置);inverse kinematics(逆运动学,已知末端位置→关节角)
📝 📝 Forward kinematics answers: "Given joint angles [30°, 45°, -20°], where is the end-effector?" — a straightforward computation using transformation matrices.
→ → 正运动学回答:"已知关节角度[30°, 45°, -20°],末端执行器在哪里?"——这是用变换矩阵的直接计算。
💡 💡 几何层面的运动描述。核心问题:已知关节角度→末端在哪?(正解);已知想到达目标位置→各关节应转多少角度?(逆解)。运动学是所有轨迹规划和控制的基础。
【中文】【中文】逆运动学 — 已知末端目标位置,求各关节需转动角度的数学过程,机器人的"逆向思维"
EN: The mathematical process of computing the joint angles required to achieve a desired end-effector position and orientation — typically involves solving nonlinear equations; computationally challenging for robots with more than 6 DOF (redundant manipulators).
🔗 🔗 简称:IK;解法:analytical(解析解,速度快但复杂机构可能无解);numerical(数值解,迭代计算,慢但通用);Damped Least Squares (DLS)(阻尼最小二乘,最常用数值法)
📝 📝 When you tell a robot "pour the water into the cup," the controller uses IK to solve for all joint angles — a nonlinear equation system that can have zero, one, or multiple valid solutions depending on robot geometry.
→ → 当你告诉机器人"把水倒进杯子里",控制器使用逆运动学解出所有关节角度——这是一个非线性方程组,根据机器人构型可能无解、有唯一解或多个解。
💡 💡 机器人的逆向思维。"我想让末端到达这里,每个关节应该转多少?"7个关节解6个自由度=数学上有无穷多解(冗余性),选择哪个解需要附加优化条件(最小能耗/最省力/最美观)。这是机器人学的数学核心难点。
【中文】【中文】轨迹规划 — 生成机器人从起点到目标点的平滑运动路径(含速度/加速度曲线)
EN: The algorithmic process of generating a time-parameterized path (position + velocity + acceleration at each instant) for the robot's end effector or joints — distinguishing from path planning (geometry only) by including temporal information.
🔗 🔗 对比:path planning(路径规划,只管几何,不管时间);trajectory(轨迹,含时间:位置+速度+加速度);jerk(加加速度,加速度的变化率);S-curve(S曲线加减速,平滑无冲击)
📝 📝 Smooth trajectory planning minimizes jerk (rate of change of acceleration) — a jerky motion during painting would cause uneven coat thickness, and during welding would accelerate electrode wear.
→ → 平滑的轨迹规划可最小化加加速度(jerk)——喷漆时冲击运动会造成涂层厚度不均,焊接时会加速焊条磨损。
💡 💡 轨迹≠路径(path)。路径只管"经过哪些点",轨迹还管"几点几分到"、"路过时多快"。冲击(jerk)是加加速度——变化太快=机械振动和寿命缩短。工业机器人用S-curve加减速取代梯形曲线(突然加速→冲击→振动)。
【中文】【中文】PID控制 — 最经典的闭环控制算法,"跟得准、稳得住、反应快"
EN: Proportional-Integral-Derivative control — a control loop feedback mechanism using error = Setpoint − Measured value: P term responds to current error, I term eliminates steady-state error, D term predicts future error from its rate of change.
🔗 🔗 变体:P(只用比例);PI(比例+积分);PD(比例+微分);调参:Ziegler-Nichols(齐格勒-尼科尔斯整定法)
📝 📝 Tuning the PID gains is essential — set P too low and the robot crawls to the target; set it too high and the robot overshoots and oscillates; the D term dampens the overshoot by sensing how fast the error is changing.
→ → 调整PID参数至关重要——P太低机器人缓慢爬向目标;P太高机器人超调并振荡;D项通过感知误差变化速度来抑制超调。
💡 💡 三个调节器:P=现在(误差×比例,直接拉动),I=过去(误差累积,消除稳态误差),D=未来(误差变化率,阻止变化太快)。调好PID,机器人"跟得准、稳得住、反应快"。虽然基础,却是工业机器人控制器中用得最多的算法。
【中文】【中文】运动规划 — 在有障碍物的环境中,让机器人从起点运动到目标而不发生碰撞的算法
EN: The computational problem of finding a sequence of valid robot configurations connecting a start state to a goal state while avoiding obstacles — typically solved in configuration space (C-space) using algorithms like RRT, PRM, or A* variants.
🔗 🔗 代表算法:RRT(快速随机探索树);PRM(概率路线图);A*(网格搜索,适合2D);RRT*(渐近最优版本)
📝 📝 In a cluttered factory floor with workers and machines, motion planning ensures the robot arm computes a collision-free path around all obstacles before executing — even when a worker steps into its workspace during operation.
→ → 在有工人和机器人的杂乱车间里,运动规划确保机器人在执行前计算出一条绕过所有障碍物的无碰撞路径——即使工人突然进入工作空间。
💡 💡 运动规划=轨迹+避障。区别于轨迹规划(已知起点终点),运动规划还要处理"障碍物在哪"。RRT(快速随机探索树)是最流行的运动规划算法,随机采样+树扩展,速度快但路径不最优(变种RRT*改善)。
【中文】【中文】示教器 — 手持编程器,操作员引导机器人记录路径点来编程
EN: A handheld industrial computer used by robot operators to manually jog (move slowly) the robot, record waypoints (teach points), program logic, and monitor status — the primary interface for offline programming and on-site debugging.
🔗 🔗 别名:teach pendant;简称:pendant;编程方式对比:online programming(示教编程);offline programming (OLP)(离线编程,在电脑上完成,不占用机器人)
📝 📝 The operator uses the teach pendant to manually jog the robot through the welding path — clicking "record" at each key waypoint — and the controller interpolates smooth motion between recorded points during playback.
→ → 操作员使用示教器手动引导机器人沿焊接路径移动——在每个关键路径点点击"记录"——控制器在回放时在记录的点之间自动插补出平滑运动。
💡 💡 "示教"=teach,顾名思义"手把手教它怎么动"。90年代工业机器人的主流编程方式——人工示教走过一遍动作,机器人记住后自动重复。缺点:占用机器人生产时间、大幅路径编程困难。优势:所见即所得,无需CAD模型。
【中文】【中文】离线编程 — 在电脑端使用CAD模型和仿真软件完成机器人编程,不占用机器人生产时间
EN: A robot programming method where path simulation, optimization, and code generation are performed on a computer using CAD models of parts and the workcell — without stopping the robot on the production line.
🔗 🔗 代表软件:RobotStudio(ABB)、RoboGuide(发那科)、KUKA.Sim(库卡)、SprutCAM(通用);对比:online teaching(在线示教,占用机器人)
📝 📝 With offline programming, a new car body variant can be programmed in 2 hours on a laptop without stopping the production line — in contrast, online teach pendant programming for the same task would require 8+ hours of robot downtime.
→ → 使用离线编程,新款车身可以在笔记本电脑上2小时完成编程,且无需停止生产线——相比之下,在线示教器编程同样任务需要8小时以上的机器人停机时间。
💡 💡 离线编程=机器人生产时的"幕后准备"。离线编程软件用CAD模型+工作单元布局进行仿真,验证轨迹是否碰撞、姿态是否可达、周期时间是否满足,然后再上传程序到真实机器人。汽车行业大规模使用OLP。
🧠 第四章 感知与人工智能(8词)
【中文】【中文】计算机视觉 — 让机器从图像/视频中提取有意义信息的科学,机器人的"眼睛"
EN: The field of enabling machines to derive meaningful information from visual inputs — encompassing image classification (what), object detection/localization (where), semantic segmentation (what exactly), and 3D reconstruction from 2D images or depth sensors.
🔗 🔗 核心任务:object detection(目标检测);semantic segmentation(语义分割);3D reconstruction(三维重建);pose estimation(位姿估计);网络:CNN(卷积神经网络);ViT(视觉Transformer)
📝 📝 Computer vision allows the robot to locate randomly placed (not fixtured) workpieces on a conveyor belt — using a depth camera to handle the third dimension and varying heights of incoming parts.
→ → 计算机视觉使机器人能够定位随机放置(无夹具固定)的工件——使用深度相机处理传入零件的第三维度和不同高度。
💡 💡 机器人的眼睛。核心任务:识别(what)、定位(where)、分割(what exactly)。技术演进:2D相机(灰度/彩色)→3D相机(深度)→事件相机(高速,捕捉像素级光变,速度比传统快1000倍)。视觉是机器人实现"随机工件抓取"(bin picking)的关键使能技术。
【中文】【中文】激光雷达 — 发射激光脉冲并分析反射来测量距离、生成3D点云的传感器
EN: Light Detection and Ranging — an active sensor that emits laser pulses and measures return time/reflection intensity to generate precise 3D point clouds of the environment — used for obstacle detection, terrain mapping, and SLAM navigation.
🔗 🔗 类型:rotating(机械旋转,360°);solid-state(固态,固态激光雷达,成本低但FOV小);波长:905nm(成本低);1550nm(人眼安全,更远距离);对比:mmWave radar(毫米波,全天候但精度低)
📝 📝 An autonomous mobile robot (AMR) in a warehouse uses LiDAR to build a real-time 3D map and localize itself within 2cm accuracy — even when pallets are moved or new shelving is installed.
→ → 仓库里的自主移动机器人(AMR)使用激光雷达构建实时3D地图并在2厘米精度内定位自身——即使托盘被移动或新货架安装后也能如此。
💡 💡 每秒发射数百万个激光点,测距精度厘米级。TOF(飞行时间)=发射→反射→接收时间×光速÷2。固态激光雷达(无机械旋转)是人形机器人的希望(体积小、成本低)。激光精确但雾天性能下降,毫米波雷达全天候但精度低——两者互补融合。
【中文】【中文】SLAM — 机器人在未知环境中边移动、边建地图、边定位自己的核心算法
EN: Simultaneous Localization and Mapping — the computational problem where a robot constructs a map of an unknown environment while simultaneously localizing itself within that map — the foundational capability for autonomous navigation.
🔗 🔗 前端:scan matching(扫描匹配);后端:pose graph optimization(位姿图优化);传感器组合:LiDAR SLAM(激光SLAM);visual SLAM(视觉SLAM,VIO);滤波:EKF SLAM(扩展卡尔曼滤波);图优化:graph SLAM(主流)
📝 📝 A warehouse robot uses SLAM to navigate dynamically — when a delivery robot moves a shelf, the robot detects the change in its LiDAR scan, updates its internal map, and re-plans a new path around the obstacle without human intervention.
→ → 仓库机器人使用SLAM动态导航——当送货机器人移动了货架,机器人检测到激光扫描的变化,更新内部地图,并自动重新规划绕过障碍物的新路径,无需人工干预。
💡 💡 "机器人在陌生房间里一边走一边画地图,同时知道自己在哪"。这是无人车、无人机、AMR的核心底层技术。SLAM = 前端(数据关联+初步估计)+ 后端(全局优化)。激光SLAM成熟稳定,视觉SLAM成本低(只用相机)但对光照敏感。
【中文】【中文】深度学习 — 用多层神经网络从海量数据中自动学习层次化表示的机器学习方法
EN: A subset of machine learning using multi-layered (deep) neural networks to automatically learn hierarchical feature representations from raw data — eliminating the need for manual feature engineering and enabling human-level accuracy on perception tasks like image classification and object detection.
🔗 🔗 网络架构:CNN(卷积神经网络,图像);RNN/LSTM(序列数据);Transformer(序列建模+视觉);GAN(生成对抗网络);diffusion model(扩散模型,图像生成)
📝 📝 Deep learning enables a robot to recognize objects with human-level accuracy — the 2012 AlexNet breakthrough (CNN, ImageNet) reduced image classification error from 26% to 15%, marking the beginning of the deep learning era in robotics vision.
→ → 深度学习使机器人能够以人类水平识别物体——2012年AlexNet突破(CNN,ImageNet)将图像分类错误率从26%降至15%,标志着机器人视觉深度学习时代的开始。
💡 💡 2012年ImageNet竞赛CNN突破是机器人视觉的转折点。深度学习=数据驱动替代人工设计特征(不再需要人工告诉机器人"边缘是重要的"),机器人从"按规则动"升级为"看过就会"。GPU并行计算使训练大规模网络成为可能。
【中文】【中文】强化学习 — 通过与环境互动、获得奖励/惩罚信号来学习最优行为策略的机器学习范式
EN: A machine learning paradigm where an agent learns optimal behavior through trial-and-error interaction with an environment, maximizing cumulative reward signals — unlike supervised learning (which requires labeled data), RL learns from the consequences of actions.
🔗 🔗 代表算法:Q-learning(值函数);DQN(深度Q网络,DeepMind Atari);PPO(近端策略优化,OpenAI,最流行);DAPG(演示辅助策略梯度);应用:sim-to-real(仿真到真实迁移)
📝 📝 Boston Dynamics trained Atlas to perform parkour using reinforcement learning in simulation — the robot fell thousands of times in simulation (with no physical damage), learned from each fall, and eventually mastered complex acrobatic maneuvers.
→ → 波士顿动力使用仿真中的强化学习训练Atlas跑酷——机器人在仿真中跌倒数千次(没有物理损坏),从每次跌倒中学习,最终掌握了复杂的杂技动作。
💡 💡 "奖励与惩罚"驱动学习——做对了奖,做错了罚。AlphaGo下棋、机器狗学走路、机器人学跑酷都是RL。优点:可学会无法用方程描述的复杂动作;缺点:样本效率极低,需要数十万甚至数百万次试错。"仿真到真实迁移"(sim-to-real)是解决样本效率问题的关键。
【中文】【中文】基座模型 — 在海量通用数据上预训练的大规模AI模型,可迁移到多种下游任务
EN: A large AI model (typically transformer-based, with billions of parameters) trained on vast and diverse datasets via self-supervised learning — then fine-tuned or prompted for specific downstream tasks (robotics, medical, language) without task-specific training from scratch.
🔗 🔗 机器人专用基座模型:RT-2(Google,VLA模型:视觉-语言-动作);PaLM-E(Google,具身多模态大模型);π₀(Physical Intelligence,2024年);OpenVLA(开源);对比:LLM(大语言模型,无视觉)
📝 📝 RT-2 (Robotic Transformer 2) is a vision-language-action (VLA) model that allows a robot to interpret commands like "push the red block toward the cup" — without any task-specific training — by leveraging general knowledge learned from web-scale data.
→ → RT-2(机器人Transformer 2)是一个视觉-语言-动作(VLA)模型,让机器人无需任何任务特定训练就能理解"把红色积木推向杯子"的命令——利用从网络规模数据中学习到的通用知识。
💡 💡 机器人领域的"GPT时刻"。预训练大模型用海量互联网数据学到"世界常识",微调后迁移到机器人任务。RT-2让机器人理解"把红色积木推向蓝色杯子"这类自然语言指令,泛化到训练中从未见过的物体和场景。2024年是VLA(视觉-语言-动作)模型的突破年。
【中文】【中文】零样本学习 — AI模型无需针对特定任务训练就能执行新任务的能力
EN: An AI model's ability to perform tasks or recognize categories it has never seen during training — by leveraging semantic knowledge from pre-training to generalize to novel situations.
📝 📝 A robot with zero-shot learning, when asked to push the blue object toward the red one, can identify and manipulate novel objects never seen in training — because its vision-language model understands color and spatial relationships from general pre-training.
→ → 使用零样本学习的机器人,当被要求把蓝色物体推向红色物体时,能识别和操作从未在训练中见过的新物体——因为它的视觉-语言模型从通用预训练中理解了颜色和空间关系。
💡 💡 零样本=从来没学过这个任务,直接做对。大模型预训练见过足够多"推东西"的场景,所以新任务也能泛化。这是让机器人从"专用"走向"通用"的关键使能技术。
【中文】【中文】仿真到真实迁移 — 在仿真中训练机器人策略,然后部署到真实机器人
EN: Training a robot policy in simulation (where data is cheap and safe) then transferring it to the physical robot — the key challenge is bridging the "reality gap" via domain randomization and system identification.
📝 📝 Boston Dynamics trained Atlas in simulation using reinforcement learning — the policy was deployed directly to the physical robot, which performed backflips on the first try without any fine-tuning on the real robot.
→ → 波士顿动力在仿真中使用强化学习训练Atlas——策略直接部署到真实机器人,机器人在第一次尝试就然后空翻,无需在真实机器人上微调。
💡 💡 真实机器人训练=慢+危险+贵;仿真=无限数据+无风险。sim-to-real是解决RL样本效率问题的工业标准路径。核心挑战:仿真物理参数和真实不完全一致(reality gap)。
🤝 第五章 人机协作(6词)
【中文】【中文】协作机器人 — 设计用于与人共享工作空间、直接交互的机器人,内置安全功能
EN: A robot specifically designed for direct human-robot interaction (HRI) within a shared workspace — featuring built-in safety systems (force limiting, collision sensing, torque monitoring) that allow it to operate without safety cages in ISO 10218 / ISO/TS 15066 compliance.
🔗 🔗 全称:collaborative robot;代表品牌:Universal Robots (UR)(全球最多装机量);发那科CR series;ABB GoFa;JAKA;AUBO
📝 📝 Cobots are deployed alongside workers for tasks requiring flexibility — a worker can reprogram the cobot in minutes by physically moving its arm to a new position (hand-guiding), without needing a teach pendant or programming expertise.
→ → 协作机器人被部署在与工人共享的工作空间中,执行需要灵活性的任务——工人可以在几分钟内通过物理引导机械臂到新位置来重新编程协作机器人,无需示教器或编程技能。
💡 💡 cobot = collaborative robot。传统工业机器人用围栏隔离(安全笼),协作机器人无围栏直接与人并肩工作。安全机制:力矩限制(碰到人立即停止)、碰撞检测(感知到接触立即降速)、手动引导(直接拖动示教)。UR机器人装机量超过5万台,是全球最多。
【中文】【中文】人机交互 — 研究人类如何与机器人沟通、协作、信任机器人的跨学科领域
EN: The interdisciplinary study of how humans communicate with, collaborate with, and build trust toward robots in shared physical or cognitive workspaces — encompassing physical HRI (touch, gesture) and cognitive HRI (language, intent recognition, UX design).
🔗 🔗 简称:HRI;对比:HMI(Human-Machine Interface,人机界面,更广义的机器界面);研究方法:user study(用户实验); Wizard of Oz(绿野仙踪,用户以为在和AI交互但有人在操控)
📝 📝 A warehouse worker can give a robot a gesture command — pointing at a shelf and nodding — and the robot interprets the intent via its vision system, autonomously navigating to pick the correct items without voice commands or touchscreen input.
→ → 仓库工人可以向机器人发出手势指令——指向货架并点头——机器人通过视觉系统解读意图,自主导航到正确位置取货,无需语音命令或触摸屏输入。
💡 💡 不只是物理交互(碰拳、共抬重物),还包括认知交互(理解人的意图、眼神、手势)。HRI研究决定人机协作的效率和人的心理接受度。协作机器人的设计必须考虑"人类信任机器人"——过度保守让人不愿用,过度激进让人害怕。
【中文】【中文】额定负载 — 末端执行器在指定条件下能可靠抓取/操作的最大重量
EN: The maximum weight a robot can reliably lift and manipulate at its end effector under specified conditions — determined by motor torque, reducer capacity, structural limits, and center-of-gravity offset from the wrist flange.
🔗 🔗 影响因素:center of gravity offset(重心偏移);inertia(惯性负载);dynamic payload(动态负载,高速运动时的负载降低系数);static payload(静态负载)
📝 📝 A 6-axis robot rated at 20 kg payload can reliably handle automotive body panels (approx. 15 kg) — but if the panel is held far from the wrist (large offset), the effective payload is reduced because the motor must resist a greater bending moment at the joint.
→ → 一款额定负载20kg的六轴机器人能可靠处理汽车车身覆盖件(约15kg)——但如果覆盖件远离腕部(大连重心偏移),有效负载会降低,因为电机必须在关节处抵抗更大的弯矩。
💡 💡 选机器人的首要参数。额定负载必须大于:夹爪重量+被抓取物重量+安全系数(通常1.2-1.5×)。超出额定负载=电机过载损坏,齿轮箱寿命急剧缩短。人形机器人常说"手部末端负载"——特斯拉Optimus目标约5-20kg。
【中文】【中文】安全速率监控 — ISO 10218-1/2定义的四种协作模式之一,人靠近时降速不停止
EN: A collaborative operation mode where the robot slows to a predefined safe speed (typically 250 mm/s) when a human enters the collaborative workspace — and resumes full speed when the human leaves. Defined in ISO 10218-1/2 and ISO/TS 15066.
🔗 🔗 四种协作模式(ISO 10218):Safety-Rated Monitored Stop (SRMS)(安全停止);Hand Guiding(手动引导);Speed & Separation Monitoring (SSM)(速度与距离监控);Power & Force Limiting (PFL)(功率与力限制,直接接触安全)
📝 📝 Under SLS, when a worker's hand enters the robot's collaborative zone (detected by safety laser scanner), the robot slows from 1m/s to 250mm/s — keeping the worker safe while maintaining production throughput, a key advantage over full stop modes.
→ → 在安全速率监控下,当工人的手进入机器人协作区域(由安全激光扫描仪检测)时,机器人从1m/s降至250mm/s——在保持工人安全的同时维持生产吞吐量,这是相比完全停止模式的关键优势。
💡 💡 ISO 10218定义的四种协作模式:SRMS(人进→停)、手动引导(人引导→机器人跟着动)、速度距离监控(人靠近→降速)、功率力限制(实时限制碰撞力)。人机协作不是"无限制共享",而是"有安全边界的共享"。
【中文】【中文】ISO机器人安全标准 — 全球工业机器人安全的基础标准,定义协作机器人的安全要求
EN: The primary international safety standards for industrial robots and robot systems — ISO 10218-1/2 covers robot machinery and integration safety; ISO/TS 15066 provides technical specifications for collaborative robot safety including force/power limits.
🔗 🔗 主要章节:ISO 10218-1(机器人本体安全);ISO 10218-2(机器人系统集成安全);ISO/TS 15066(协作机器人技术规范);地区版本:ANSI/RIA R15.06(美国);GB/T 16855.1(中国)
📝 📝 Under ISO/TS 15066, the maximum permissible quasi-static contact force on a human body part is 130 N for the chest and 65 N for the face — cobots must be designed to limit collision forces below these thresholds.
→ → 根据ISO/TS 15066,胸部最大允许准静态接触力为130N,面部为65N——协作机器人必须设计为将碰撞力限制在这些阈值以下。
💡 💡 ISO 10218是工业机器人安全的"宪法"。任何进入欧洲市场的工业机器人必须CE认证(基于ISO 10218)。ISO/TS 15066补充了协作机器人具体的技术参数,包括身体各部位允许的最大接触力。安全是工业机器人的底线。
【中文】【中文】手动引导 — 操作员直接拖动机器人手臂示教,无需示教器,协作机器人特有功能
EN: A collaborative robot programming method where the operator physically moves the robot arm through the desired trajectory — the robot records joint angles; gravity compensation makes the arm feel almost weightless during guidance.
📝 📝 With hand guiding, a factory worker without programming experience can reprogram a cobot in 3 minutes — by simply grabbing and moving the arm to show the desired path, then tapping the wrist button to save each waypoint.
→ → 有了手动引导,没有编程经验的工人可以在3分钟内重新编程协作机器人——只需抓住手臂移动到所需轨迹,然后点击腕部按钮保存路径点。
💡 💡 协作机器人的核心优势之一。操作员直接拖动,机器人提供重力补偿(感觉像浮在水里)=编程门槛降到零。UR机器人能够普及到中小企业的关键原因之一。
🦾 第六章 构型与分类(9词)
【中文】【中文】关节型机器人 — 串联关节结构的旋转关节机械臂,最常见的工业机器人类型
EN: A robot with rotary joints (typically 4 to 6 axes arranged in series, resembling a human arm structure) — the dominant industrial robot type accounting for ~60% of global installations, used for welding, painting, assembly, and material handling.
🔗 🔗 别名:serial robot(串联机器人);常见配置:6-axis(最多),4-axis(SCARA变种),7-axis冗余(多一个自由度用于奇异规避);四大家族:发那科/ABB/库卡/安川
📝 📝 Articulated robots dominate automotive assembly lines — their 6-axis serial structure provides the flexibility to reach car body weld points from multiple angles without repositioning the heavy workpiece, reducing cycle time by 40% vs. fixed automation.
→ → 关节型机器人主导着汽车装配线——其6轴串联结构能从多角度到达白车身焊点而无需重新定位重型工件,与固定自动化相比将周期时间缩短了40%。
💡 💡 6轴=人手臂自由度,可到达空间任意位置和姿态。汽车点焊、喷涂、装配的主力机型。发那科FANUC、ABB IRB、库卡KR、安川MOTOMAN,四大家族全部以关节型为核心。结构:底座固定→肩部(轴1)→上臂(轴2)→前臂(轴3)→腕部(轴4-6)。
【中文】【中文】SCARA机器人 — 四轴平行关节结构,专攻垂直方向的取放和装配,高速高精度
EN: Selective Compliance Assembly Robot Arm — a 4-axis robot configuration with two parallel rotary joints (horizontal plane) plus a vertical prismatic joint — optimized for fast vertical pick-and-place and insertion tasks in electronics, pharmaceuticals, and consumer goods assembly.
🔗 🔗 全称:Selective Compliance Assembly Robot Arm;特点:圆柱形工作空间,只能垂直上下+水平移动,不能倾斜末端姿态;代表:EPSON ForcePro / Canon Robot
📝 📝 SCARA robots dominate electronics assembly because they achieve cycle times under 0.4 seconds for pick-and-place — critical for placing 0201 resistors on PCBs at 10,000 units/hour on SMT lines.
→ → SCARA机器人因其取放周期时间低于0.4秒而在电子装配中占主导地位——这对于在SMT生产线上以每小时10000件的速度将0201电阻放置到PCB上至关重要。
💡 💡 4轴(X/Y/Z+旋转),只能在平面内大范围移动和垂直上下,不能"歪着拿东西"(无法绕水平轴倾斜)。主要用于电子/半导体装配,速度快(周期0.3-0.5秒)、精度高(±0.02mm)、价格低。爱普生是全球SCARA出货量第一。
【中文】【中文】Delta机器人 — 并联三臂结构,运动部件极轻的超高速分拣机器人
EN: A parallel kinematic robot with three identical arms connected to a common base, forming a triangular kinematic chain — because the motors are fixed to the base (not the moving platform), only the lightweight arms move, enabling extremely high speed pick-and-place (up to 300+ picks/min).
🔗 🔗 别名:parallel manipulator(并联机械臂);优势:high speed(高速,200-300次/分);low inertia(低惯性);high stiffness(高刚性);局限:small work envelope(工作空间有限)
📝 📝 Delta robots operating at 200+ picks per minute sort and pack chocolates into gift boxes — the lightweight moving platform (only arms move, motors stay on fixed base) enables accelerations of 30g without damaging the delicate products.
→ → 每分钟200次以上抓取的Delta机器人在巧克力分拣包装中运行——轻量化的移动平台(只有臂在动,电机固定在底座)实现了30g的加速度而不损坏精美产品。
💡 💡 底部固定,三臂向下一路并联合并到末端执行器,像倒置的金字塔。运动部件极轻(只有执行器在动,电机全部固定在顶部落支架上)=惯性极低=速度极快。典型应用:食品/药品/化妆品高速分拣包装(每分钟200-300次)。工作空间是倒金字塔形,比较小。
【中文】【中文】AGV — 沿固定物理路径(磁条/电线/光学)行驶的自动运输车,"看不见路的搬运工"
EN: Automated Guided Vehicle — a mobile robot that follows predetermined fixed routes using physical infrastructure (magnetic tape embedded in floor, buried wire, or optical floor markers) to transport pallets, carts, or raw materials in factories and warehouses without real-time obstacle avoidance.
🔗 🔗 导引方式:magnetic tape(磁条);wire-guided(地磁线);optical(光学);laser navigation(激光信标);对比:AMR(自主移动机器人,可实时避障)
📝 📝 AGVs have been used in automotive assembly plants since the 1980s to transport chassis — they follow a single fixed route (embedded magnetic tape) reliably but require manual reconfiguration of the floor whenever the production layout changes.
→ → 自1980年代以来,AGV一直被用于汽车装配工厂运输底盘——它们沿固定路线(埋地磁条)可靠行驶,但每当生产布局改变时就需要手动重新配置地面。
💡 💡 沿固定路线走(磁条/电线/光学),不能自主绕障,路线变更需改物理基础设施。优点:成熟可靠;缺点:不灵活。适合单一品种大批量物流(汽车装配线)。与AMR的核心区别:AGV"看不见",AMR"看得见"能绕障。
【中文】【中文】AMR — 用传感器和AI实时感知环境、动态绕障的自主移动机器人
EN: Autonomous Mobile Robot — a mobile robot using onboard sensors (LiDAR, cameras) and AI to navigate dynamically, detecting and avoiding obstacles and adapting routes in real time — unlike AGVs, AMRs do not require fixed infrastructure and can operate in environments with unpredictable human movement.
🔗 🔗 代表品牌:亚马逊Kiva/Multi-Bot;极智嘉Geek+(中国);海康机器人;Fetch Robotics(Fetch+Freight);Otomate(世仓)
📝 📝 Unlike AGVs, AMRs do not require warehouse floor modification — they use LiDAR + SLAM to autonomously map their environment on the first run, then navigate around temporary obstacles like forklifts and stacked pallets without stopping.
→ → 与AGV不同,AMR不需要改造仓库设施——它们使用激光雷达+SLAM在首次运行时自主建图,然后在叉车和堆叠托盘等临时障碍物周围导航而不停止。
💡 💡 AMR = AGV的升级版。AGV"看不见"只能走固定路,AMR有视觉/LiDAR"长眼睛"能绕障。极智嘉(Geek+)是中国AMR龙头,亚马逊Kiva是全球电商仓储AMR的鼻祖。核心差异:AMR不改造地面,实时建图,动态路径规划。
【中文】【中文】人形机器人 — 模拟人类身体形态(双臂+双腿+头部)的机器人,终极目标是无需改造人类环境
EN: A robot whose physical form resembles the human body — typically a torso, two arms with hands (including articulated fingers), a head with sensors, and sometimes legs — designed to operate in human environments, use human tools, and navigate spaces designed for people.
🔗 🔗 关键指标:degrees of freedom(手部27 DOF=全身60+ DOF);平衡控制:ZMP(零力矩点);WBC(全身协调控制);驱动方式:electric(电动,主流);hydraulic(液压,Atlas,力量型)
📝 📝 Tesla Optimus is designed to navigate human environments using the same spaces designed for people — stairs, door handles, standard tables — without modifications, using its whole-body coordination (WBC) to maintain balance while carrying a 20 kg load up stairs.
→ → 特斯拉Optimus被设计用来在使用为人类设计的环境中导航——楼梯、门把手、标准桌子——无需改造,它使用全身协调控制(WBC)在携带20kg负载上楼时保持平衡。
💡 💡 终极目标:让机器人无需改造人类环境即可工作。双足平衡(ZMP/WBC算法)+灵巧手(27个自由度)+全身协调是三大技术门槛。全身60+自由度,远超工业机器人的6 DOF,需要实时全身协调控制。2025年是多家人形机器人量产的起点。
【中文】【中文】外骨骼 — 穿戴在人身上的机器人,增强而非替代人类运动能力
EN: A wearable robotic device worn by a human that augments, assists, or enables movement of the user's limbs by applying mechanical forces — unlike humanoid robots (which replace humans), exoskeletons enhance human capability; applications in rehabilitation medicine, industrial overhead work, and military load-carrying.
🔗 🔗 类型:powered exoskeleton(电动/液压驱动);passive exoskeleton(无源,纯机械助力);应用:industrial(减少工人疲劳);medical/rehab(中风康复);military(增加士兵负载能力);代表:Cyberdyne HAL(日本,FDA批准)
📝 📝 An industrial exoskeleton allows a factory worker to perform overhead tasks for 8 hours without shoulder fatigue — the exoskeleton transfers the arm's weight to the hips, reducing shoulder muscle load by 40% during screwdriver assembly operations.
→ → 工业外骨骼使工厂工人能够连续8小时执行头顶作业而不感到肩部疲劳——外骨骼将手臂重量转移到髋部,在螺丝刀装配作业中将肩部肌肉负荷减少40%。
💡 💡 外骨骼=穿在身上的机器人。与人形机器人相反:人形替代人,外骨骼增强人(人在系统里)。三大方向:医疗康复(中风后步行训练)、工业重载(减少工人疲劳)、军事(增强士兵负载)。Cyberdyne HAL是日本外骨骼龙头,FDA已批准用于医疗。
【中文】【中文】拟人化机器人 — 具有人类外观特征(人形面孔/表情/头部)的服务或社交机器人
EN: A robot designed to appear or behave in a human-like manner — typically featuring a humanoid face, eyes, facial expressions, or conversational ability — used in customer service, healthcare companionship, and research into human-robot interaction.
🔗 🔗 对比:humanoid(人形,身体结构类似人,Motor可以不同);android(安卓,外貌完全模仿人类);gynoid(女性外观的拟人机器人);代表:Sophia(Hanson Robotics);Ameca(Engineered Arts)
📝 📝 Sophia the robot, developed by Hanson Robotics, uses generative facial expressions — her AI interprets conversational context and synthesizes 60+ facial muscle configurations in real time to match human emotional responses, creating the illusion of empathy during healthcare interactions.
→ → 汉森机器人公司开发的索菲亚机器人使用生成式面部表情——她的AI实时解读对话语境并综合60多种面部肌肉配置以匹配人类情感反应,在医疗交互中创造共情错觉。
💡 💡 拟人化≠人形(humanoid)。拟人化强调外观/行为的"像人",与人形机器人的"身体结构像人"有区别。拟人化设计在人机交互场景(医疗、接待)能提高人类接受度,但也带来"恐怖谷"效应——太逼真但不够逼真会让人不适。
【中文】【中文】群体机器人学 — 研究大量简单机器人通过局部规则实现群体智能的学科
EN: A branch of robotics studying how large groups of relatively simple, homogeneous robots can collectively accomplish complex tasks through local rules and emergent behavior — inspired by social insects (ants, bees, termites) that achieve sophisticated construction and foraging without centralized control.
🔗 🔗 核心概念:emergence(涌现,群体行为≠个体行为的简单加总);stigmergy(蜂群智能,通过环境间接通信);算法:粒子群优化(PSO);应用:warehouse logistics(亚马逊Kiva);agriculture(群体采摘);construction(群体3D打印)
📝 📝 Amazon warehouses use swarm robotics — when you order a book, the nearest available robot navigates to its shelf location (self-organized without a central dispatcher), picks up the pod, and delivers it to the human picker — 10,000 robots operating simultaneously without centralized coordination.
→ → 亚马逊仓库使用群体机器人技术——当你订购一本书时,最近的可用机器人自主导航到其货架位置(无中央调度器的自组织),拾取货架并将其运送给人类拣货员——10000个机器人同时运行而无需中央协调。
💡 💡 灵感来自蚂蚁/蜜蜂:每个个体很简单(只能感知局部),但群体行为涌现出惊人智能(建造复杂巢穴、找到最短路径)。优势:去中心化=无单点故障;可扩展=加机器人即扩容;容错=几个坏了整体继续工作。
🚀 第七章 前沿技术(6词)
【中文】【中文】数字孪生 — 物理机器人/系统的实时虚拟副本,用于仿真、监控和预测性维护
EN: A high-fidelity virtual replica of a physical robot, production cell, or factory system — continuously synchronized with the real asset via sensor data — used for offline simulation (reducing real-world debugging time), predictive maintenance (detecting bearing wear from vibration), and remote monitoring.
🔗 🔗 层级:component twin(单关节);system twin(工作单元);fleet twin(多机器人车队);代表平台:Siemens Plant Simulation;ABB RobotStudio;PTC ThingWorx
📝 📝 Engineers use a digital twin to simulate a new robot program for 100 cycles on the virtual robot before downloading it to the real robot — identifying collision risks and optimizing cycle time in simulation, reducing real commissioning time from 3 days to 4 hours.
→ → 工程师使用数字孪生在虚拟机器人上仿真新程序运行100个周期,然后再下载到真实机器人——在仿真中识别碰撞风险并优化周期时间,将实际调试时间从3天缩短到4小时。
💡 💡 物理机器人的"数字克隆体"。传感器实时同步数据,虚实双向映射。数字孪生=仿真(调试)+监控(运行状态)+预测(故障预警)。ABB RobotStudio和FANUC RoboGuide是最流行的机器人数字孪生平台。
【中文】【中文】工业4.0 — 融合物联网、AI、云计算与机器人的智能制造趋势
EN: The ongoing fourth industrial revolution in manufacturing — integrating Cyber-Physical Systems (CPS), IoT sensors, AI analytics, cloud computing, and networked robots via standard protocols (OPC-UA, MQTT) to achieve intelligent, self-optimizing production with full digital traceability.
🔗 🔗 核心概念:CPS(信息物理系统);IIoT(工业物联网);OPC-UA(机器设备间安全数据交换标准);MES(制造执行系统);演进:Industry 3.0(自动化单机)→Industry 4.0(智能网络化)
📝 📝 An Industry 4.0 factory features networked robots communicating via OPC-UA — each robot reports cycle time, tool wear, and energy consumption in real time to the MES, which uses AI to predict quality defects and autonomously adjust process parameters 30 minutes before they occur.
→ → 工业4.0工厂的特色是网络化机器人通过OPC-UA通信——每台机器人实时向MES报告周期时间、刀具磨损和能耗,MES使用AI预测质量缺陷并在缺陷发生前30分钟自动调整工艺参数。
💡 💡 工业1.0(蒸汽机)→2.0(电)→3.0(电子/IT,PLC和工业电脑)→4.0(智能网络)。核心是CPS(信息物理系统)+MES(制造执行系统)+机器人无缝连接。目标是原材料进厂到成品出厂全程数据驱动、实时优化、无人干预。
【中文】【中文】边缘计算 — 在数据产生源头(机器人/传感器)附近进行的计算,无需云端往返
EN: Computing infrastructure deployed near the point of data generation (robot, sensor, or edge gateway) to enable real-time AI inference, sub-10ms latency, and operation without cloud connectivity — the "local brain" of an industrial robot or autonomous vehicle.
🔗 🔗 对比:cloud computing(云端,延迟高但算力强);fog computing(雾计算,介于边缘和云之间);典型硬件:NVIDIA Jetson AGX(边缘AI平台);Intel Movidius VPU(视觉处理单元);Google Edge TPU(张量处理单元)
💡 💡 机器人要求毫秒级响应(协作碰撞检测<10ms),云端往返延迟不可接受。边缘计算=在机器人身边放个小电脑(NVIDIA Jetson/Intel Movidius)实时推理。工厂内机器人大多在边缘端运行,视觉推理是边缘AI最大的算力消耗。Jetson Orin Nano能跑30 TOPS,足够处理视觉模型。
【中文】【中文】遥操作 — 操作员在远程控制机器人,延迟是核心挑战
EN: Remote control of a robot by a human operator not physically present — using video feed, force feedback, and control signals over a network; bilateral control provides force feedback; latency is the primary challenge for fine manipulation.
🔗 🔗 应用:达芬奇手术机器人(Intuitive Surgical);危险环境(核电站/拆弹);关键技术:bilateral control(双向力反馈);round-trip time(往返时延)
📝 📝 A surgeon in New York performs surgery on a patient in Tokyo using the Da Vinci robot — viewing 3D HD video with 10x magnification and controlling micro-instruments with sub-millimeter precision, compensating for 200ms latency with predictive display algorithms.
→ → 纽约外科医生使用达芬奇机器人为东京患者手术——以10倍放大3D高清视频观察,用手部动作控制亚毫米精度微型器械,使用预测显示算法补偿200ms网络延迟。
💡 💡 遥操作=人-机-网络-机-人闭环。超过200ms延迟后手眼协调被破坏,精细操作几乎不可能。达芬奇手术机器人已完成超1000万例。
【中文】【中文】全身协调控制 — 同时协调人形机器人所有关节(60+ DOF)实现全身运动
EN: A control architecture simultaneously coordinating all joints of a complex robot (60+ DOF for humanoid) — treating the entire body as a unified system rather than independent limb controllers.
📝 📝 Using whole-body control, a humanoid robot walks while balancing a tray — the controller coordinates legs for gait, torso for balance, and arms for tray orientation simultaneously, solved in a single QP at 1kHz.
→ → 使用全身协调控制,人形机器人可以边走边平衡托盘——控制器同时协调腿部步态、躯干平衡和手臂托盘方向,全部在1kHz的二次规划中求解。
💡 💡 工业机械臂只控6 DOF,问题简单。人形60+ DOF=同时解60+个方程,必须实时(1kHz=1毫秒内解完)。WBC是人形机器人控制的核心算法。
【中文】【中文】混合智能 — 人类智能与AI智能深度融合,AI处理数据,人把控方向
EN: A paradigm where human and machine intelligence are tightly integrated — the human provides high-level intent and ethical judgment; the AI provides speed, precision, and tireless data processing, with each correcting the other in a continuous loop.
📝 📝 In hybrid intelligence, a quality inspector marks defects on X-ray images and the AI learns from each correction — the AI autonomously catches 95% of defects over time, and the inspector reviews only the 5% uncertain cases, reducing inspection time by 10x.
→ → 在混合智能模式下,质量检查员在铸件X射线图像上标记缺陷,AI从每次纠正中学习——一段时间后AI自主捕获95%的缺陷,检查员只需审核5%的不确定案例,将检查时间减少10倍,同时保持质量。
💡 💡 纯AI=快但缺上下文判断;纯人类=准但慢且贵。混合智能=AI做重复性工作,人做创意性审核。AI从每次纠正中学习,形成持续改进闭环。这是当前工业AI落地的最务实路径。
↑