代码拉取完成，页面将自动刷新
同步操作将从 eric/underactuated-2020 强制同步，此操作会覆盖自 Fork 仓库以来所做的任何修改，且无法恢复！！！
确定后同步将在后台操作，完成时将刷新页面，请耐心等待。
<!DOCTYPE html>
<html>
<head>
<title>Underactuated Robotics</title>
<meta name="Underactuated Robotics" content="text/html; charset=utf-8;" />
<script type="text/javascript" src="chapters.js"></script>
<script type="text/javascript" src="htmlbook/book.js"></script>
<script src="htmlbook/mathjax-config.js" defer></script>
<script type="text/javascript" id="MathJax-script" defer
src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-svg.js">
</script>
<script>window.MathJax || document.write('<script type="text/javascript" src="htmlbook/MathJax/es5/tex-svg.js" defer><\/script>')</script>
<link rel="stylesheet" href="htmlbook/highlight/styles/default.css">
<script src="htmlbook/highlight/highlight.pack.js"></script> <!-- http://highlightjs.readthedocs.io/en/latest/css-classes-reference.html#language-names-and-aliases -->
<script>hljs.initHighlightingOnLoad();</script>
<link rel="stylesheet" type="text/css" href="htmlbook/book.css">
</head>
<body onload="forwardOldChapterLink(); customTags(); MathJax.typeset();">
<div data-type="titlepage">
<header>
<h1><a href="index.html" style="text-decoration:none;">Underactuated Robotics</a></h1>
<p data-type="subtitle">Algorithms for Walking, Running, Swimming, Flying, and Manipulation</p>
<p style="font-size: 18px;"><a href="http://people.csail.mit.edu/russt/">Russ Tedrake</a></p>
<p style="font-size: 14px; text-align: right;">
© Russ Tedrake, 2020<br/>
<a href="tocite.html">How to cite these notes</a> |
<a target="_blank" href="https://docs.google.com/forms/d/e/1FAIpQLSesAhROfLRfexrRFebHWLtRpjhqtb8k_iEagWMkvc7xau08iQ/viewform?usp=sf_link">Send me your feedback</a><br/>
</p>
</header>
</div>
<p><b>Note:</b> These are working notes used for <a
href="http://underactuated.csail.mit.edu/Spring2020/">a course being taught
at MIT</a>. They will be updated throughout the Spring 2020 semester. <a
href="https://www.youtube.com/channel/UChfUOAhz7ynELF-s_1LPpWg">Lecture videos are available on YouTube</a>.</p>
<section id="table_of_contents">
<h1>Table of Contents</h1>
<ul>
<li><a href="#preface">Preface</a></li>
<li><a href="intro.html">Chapter 1: Fully-actuated vs Underactuated Systems</a></li>
<ul>
<li><a href=intro.html#section1>Motivation</a></li>
<ul>
<li>Honda's ASIMO vs. passive dynamic walkers</li>
<li>Birds vs. modern aircraft</li>
<li>Manipulation</li>
<li>The common theme</li>
</ul>
<li><a href=intro.html#section2>Definitions</a></li>
<li><a href=intro.html#section3>Feedback Equivalence</a></li>
<li><a href=intro.html#section4>Input and State Constraints</a></li>
<ul>
<li>Nonholonomic constraints</li>
</ul>
<li><a href=intro.html#section5>Underactuated robotics</a></li>
<li><a href=intro.html#section6>Goals for the course</a></li>
<li><a href=intro.html#section7>Exercises</a></li>
</ul>
<p style="margin-bottom: 0; text-decoration: underline;font-variant: small-caps;"><b>Model Systems</b></p>
<li><a href="pend.html">Chapter 2: The Simple Pendulum</a></li>
<ul>
<li><a href=pend.html#section1>Introduction</a></li>
<li><a href=pend.html#section2>Nonlinear dynamics with a
constant
torque</a></li>
<ul>
<li>The overdamped pendulum</li>
<li>The undamped pendulum with zero torque</li>
<li>The undamped pendulum with a constant
torque</li>
</ul>
<li><a href=pend.html#section3>The torque-limited simple pendulum</a></li>
<ul>
<li>Energy-shaping control</li>
</ul>
<li><a href=pend.html#section4>Exercises</a></li>
</ul>
<li><a href="acrobot.html">Chapter 3: Acrobots, Cart-Poles, and Quadrotors</a></li>
<ul>
<li><a href=acrobot.html#section1>The Acrobot</a></li>
<ul>
<li>Equations of motion</li>
</ul>
<li><a href=acrobot.html#cart_pole>The Cart-Pole system</a></li>
<ul>
<li>Equations of motion</li>
</ul>
<li><a href=acrobot.html#section3>Quadrotors</a></li>
<ul>
<li>The Planar Quadrotor</li>
<li>The Full 3D Quadrotor</li>
</ul>
<li><a href=acrobot.html#section4>Balancing</a></li>
<ul>
<li>Linearizing the manipulator equations</li>
<li>Controllability of linear systems</li>
<li>LQR feedback</li>
</ul>
<li><a href=acrobot.html#section5>Partial feedback linearization</a></li>
<ul>
<li>PFL for the Cart-Pole System</li>
<li>General form</li>
</ul>
<li><a href=acrobot.html#section6>Swing-up control</a></li>
<ul>
<li>Energy shaping</li>
<li>Cart-Pole</li>
<li>Acrobot</li>
<li>Discussion</li>
</ul>
<li><a href=acrobot.html#differential_flatness>Differential Flatness</a></li>
<li><a href=acrobot.html#section8>Other model systems</a></li>
<li><a href=acrobot.html#section9>Exercises</a></li>
</ul>
<li><a href="simple_legs.html">Chapter 4: Simple Models
of Walking and Running</a></li>
<ul>
<li><a href=simple_legs.html#section1>Limit Cycles</a></li>
<ul>
<li>Poincaré Maps</li>
</ul>
<li><a href=simple_legs.html#section2>Simple Models of Walking</a></li>
<ul>
<li>The Rimless Wheel</li>
<li>The Compass Gait</li>
<li>The Kneed Walker</li>
<li>Curved feet</li>
<li>And beyond...</li>
</ul>
<li><a href=simple_legs.html#section3>Simple Models of Running</a></li>
<ul>
<li>The Spring-Loaded Inverted Pendulum</li>
<li>Hopping Robots from the MIT Leg Laboratory</li>
</ul>
<li><a href=simple_legs.html#section4>A Simple Model That Can Walk and Run</a></li>
<li><a href=simple_legs.html#section5>Exercises</a></li>
</ul>
<li><a href="humanoids.html">Chapter 5: Highly-articulated Legged
Robots</a></li>
<ul>
<li><a href=humanoids.html#section1>Center of Mass Dynamics</a></li>
<ul>
<li>A hovercraft model</li>
<li>Robots with (massless) legs</li>
<li>Capturing the full robot dynamics</li>
<li>Impact dynamics</li>
</ul>
<li><a href=humanoids.html#section2>The special case of flat terrain</a></li>
<ul>
<li>An aside: the zero-moment point
derivation</li>
</ul>
<li><a href=humanoids.html#section3>ZMP planning</a></li>
<ul>
<li>From a CoM plan to a whole-body
plan</li>
</ul>
<li><a href=humanoids.html#section4>Whole-Body Control</a></li>
<li><a href=humanoids.html#section5>Footstep planning and push recovery</a></li>
<li><a href=humanoids.html#section6>Beyond ZMP planning</a></li>
<li><a href=humanoids.html#section7>Exercises</a></li>
</ul>
<li><a href="stochastic.html">Chapter 6: Model Systems
with Stochasticity</a></li>
<ul>
<li><a href=stochastic.html#section1>The Master Equation</a></li>
<li><a href=stochastic.html#section2>Stationary Distributions</a></li>
<li><a href=stochastic.html#section3>Extended Example: The Rimless Wheel on Rough
Terrain</a></li>
<li><a href=stochastic.html#section4>Noise models for real robots/systems.</a></li>
<li><a href=stochastic.html#section5>Analysis (a preview)</a></li>
<li><a href=stochastic.html#section6>Control Design (a preview)</a></li>
</ul>
<p style="margin-bottom: 0; text-decoration: underline;font-variant: small-caps;"><b>Nonlinear Planning and Control</b></p>
<li><a href="dp.html">Chapter 7: Dynamic Programming</a></li>
<ul>
<li><a href=dp.html#section1>Formulating control design as an optimization</a></li>
<ul>
<li>Additive cost</li>
</ul>
<li><a href=dp.html#section2>Optimal control as graph search</a></li>
<li><a href=dp.html#section3>Continuous dynamic programming</a></li>
<ul>
<li>The Hamilton-Jacobi-Bellman Equation</li>
<li>Solving for the minimizing control</li>
<li>Numerical solutions for $J^*$</li>
</ul>
<li><a href=dp.html#section4>Extensions</a></li>
<ul>
<li>Linear Programming Approach</li>
</ul>
<li><a href=dp.html#section5>Exercises</a></li>
</ul>
<li><a href="lqr.html">Chapter 8: Linear Quadratic Regulators</a></li>
<ul>
<li><a href=lqr.html#section1>Basic Derivation</a></li>
<ul>
<li>Local stabilization of nonlinear systems</li>
</ul>
<li><a href=lqr.html#finite_horizon>Finite-horizon formulations</a></li>
<ul>
<li>Finite-horizon LQR</li>
<li>Time-varying LQR</li>
<li>Linear Quadratic Optimal Tracking</li>
<li>Linear Final Boundary Value Problems</li>
</ul>
<li><a href=lqr.html#section3>Variations and extensions</a></li>
<ul>
<li>Discrete-time Riccati Equations</li>
<li>LQR with input and state constraints</li>
<li>LQR as a convex optimization</li>
<li>Finite-horizon LQR via least
squares</li>
</ul>
<li><a href=lqr.html#section4>Exercises</a></li>
<li><a href=lqr.html#section5>Notes</a></li>
<ul>
<li>Finite-horizon LQR derivation (general form)</li>
</ul>
</ul>
<li><a href="lyapunov.html">Chapter 9: Lyapunov
Analysis</a></li>
<ul>
<li><a href=lyapunov.html#section1>Lyapunov Functions</a></li>
<ul>
<li>Global Stability</li>
<li>LaSalle's Invariance Principle</li>
<li>Relationship to the Hamilton-Jacobi-Bellman
equations</li>
</ul>
<li><a href=lyapunov.html#section2>Lyapunov analysis with convex optimization</a></li>
<ul>
<li>Lyapunov analysis for linear systems</li>
<li>Lyapunov analysis as a semi-definite program
(SDP)</li>
<li>Lyapunov analysis for polynomial systems</li>
</ul>
<li><a href=lyapunov.html#section3>Lyapunov functions for estimating regions of
attraction</a></li>
<ul>
<li>Robustness analysis using "common Lyapunov functions"</li>
<li>Region of attraction estimation for polynomial systems</li>
</ul>
<li><a href=lyapunov.html#section4>Rigid-body dynamics are (rational) polynomial</a></li>
<li><a href=lyapunov.html#section5>Exercises</a></li>
</ul>
<li><a href="trajopt.html">Chapter 10: Trajectory
Optimization</a></li>
<ul>
<li><a href=trajopt.html#section1>Problem Formulation</a></li>
<li><a href=trajopt.html#section2>Convex Formulations for Linear Systems</a></li>
<ul>
<li>Direct Transcription</li>
<li>Direct Shooting</li>
<li>Computational Considerations</li>
<li>Continuous Time</li>
</ul>
<li><a href=trajopt.html#section3>Nonconvex Trajectory Optimization</a></li>
<ul>
<li>Direct Transcription and Direct Shooting</li>
<li>Direct Collocation</li>
<li>Pseudo-spectral Methods</li>
<li>Solution techniques</li>
</ul>
<li><a href=trajopt.html#section4>Local Trajectory Feedback Design</a></li>
<ul>
<li>Time-varying LQR</li>
<li>Model-Predictive Control</li>
</ul>
<li><a href=trajopt.html#perching>Case Study: A glider that can land on a perch
like a bird</a></li>
<ul>
<li>The Flat-Plate Glider Model</li>
<li>Trajectory optimization</li>
<li>Trajectory stabilization</li>
<li>Beyond a single trajectory</li>
</ul>
<li><a href=trajopt.html#section6>Pontryagin's Minimum Principle</a></li>
<ul>
<li>Lagrange multiplier derivation of the adjoint equations</li>
<li>Necessary conditions for optimality in continuous time</li>
</ul>
<li><a href=trajopt.html#section7>Variations and Extensions</a></li>
<ul>
<li>Iterative LQR and Differential Dynamic
Programming</li>
<li>Mixed-integer convex optimization for non-convex
constraints</li>
<li>Explicit model-predictive control</li>
</ul>
<li><a href=trajopt.html#section8>Exercises</a></li>
</ul>
<li><a href="planning.html">Chapter 11: Motion Planning as
Search</a></li>
<ul>
<li><a href=planning.html#section1>Artificial Intelligence as Search</a></li>
<li><a href=planning.html#section2>Randomized motion planning</a></li>
<ul>
<li>Rapidly-Exploring Random Trees (RRTs)</li>
<li>RRTs for robots with dynamics</li>
<li>Variations and extensions</li>
<li>Discussion</li>
</ul>
<li><a href=planning.html#section3>Decomposition methods</a></li>
<li><a href=planning.html#section4>Exercises</a></li>
</ul>
<li><a href="feedback_motion_planning.html">Chapter 12: Feedback Motion
Planning</a></li>
<li><a href="policy_search.html">Chapter 13: Policy
Search</a></li>
<ul>
<li><a href=policy_search.html#section1>Problem formulation</a></li>
<li><a href=policy_search.html#section2>Controller parameterizations</a></li>
<li><a href=policy_search.html#section3>Trajectory-based policy search</a></li>
<li><a href=policy_search.html#section4>Lyapunov-based approaches to policy
search.</a></li>
<li><a href=policy_search.html#section5>Approximate Dynamic Programming</a></li>
</ul>
<li><a href="robust.html">Chapter 14: Robust and
Stochastic Control</a></li>
<ul>
<li><a href=robust.html#section1>Discrete states and actions</a></li>
<ul>
<li>Graph search -- stochastic shortest path
problems</li>
<li>Stochastic interpretation of deterministic,
continuous-state value iteration</li>
</ul>
<li><a href=robust.html#section2>Linear Quadratic Gaussian (LQG)</a></li>
<li><a href=robust.html#section3>Linear Exponential-Quadratic Gaussian (LEQG)</a></li>
</ul>
<li><a href="output_feedback.html">Chapter 15:
Output Feedback (aka Pixels-to-Torques)</a></li>
<ul>
<li><a href=output_feedback.html#section1>The Classical Perspective</a></li>
<li><a href=output_feedback.html#section2>Observer-based Feedback</a></li>
<ul>
<li>Luenberger Observer</li>
<li>Linear Quadratic Regulator w/ Gaussian Noise
(LQG)</li>
<li>Partially-observable Markov Decision Processes</li>
</ul>
<li><a href=output_feedback.html#section3>Static Output Feedback</a></li>
<ul>
<li>For Linear Systems</li>
</ul>
<li><a href=output_feedback.html#section4>Disturbance-based feedback</a></li>
<ul>
<li>System-Level Synthesis</li>
</ul>
</ul>
<li><a href="limit_cycles.html">Chapter 16: Algorithms
for Limit Cycles</a></li>
<ul>
<li><a href=limit_cycles.html#trajopt>Trajectory optimization</a></li>
<li><a href=limit_cycles.html#lyapunov>Lyapunov analysis</a></li>
<ul>
<li>Transverse coordinates</li>
<li>Transverse linearization</li>
<li>Region of attraction estimation using sums-of-squares</li>
</ul>
<li><a href=limit_cycles.html#section3>Feedback design</a></li>
<ul>
<li>For underactuation degree one.</li>
<li>Transverse LQR</li>
<li>Orbital stabilization for non-periodic trajectories</li>
</ul>
</ul>
<li><a href="contact.html">Chapter 17: Planning and
Control through Contact</a></li>
<ul>
<li><a href=contact.html#section1>(Autonomous) Hybrid Systems</a></li>
<ul>
<li>Hybrid trajectory optimization</li>
<li>Stabilizing hybrid models.</li>
<li>Deriving hybrid models: minimal vs floating-base
coordinates</li>
</ul>
</ul>
<p style="margin-bottom: 0; text-decoration: underline;font-variant: small-caps;"><b>Estimation and Learning</b></p>
<li><a href="sysid.html">Chapter 18: System Identification</a></li>
<ul>
<li><a href=sysid.html#section1>Equation Error vs Simulation Error</a></li>
<li><a href=sysid.html#section2>Linear State-Space Models</a></li>
<ul>
<li>From state observations</li>
<li>From input-output data</li>
<li>Adding stability constraints</li>
</ul>
<li><a href=sysid.html#section3>Linear ARMAX Models</a></li>
<li><a href=sysid.html#section4>Nonlinear System Identification</a></li>
<li><a href=sysid.html#section5>Parameter Identification for Mechanical Systems</a></li>
<li><a href=sysid.html#section6>Optimal Experiment Design</a></li>
</ul>
<li><a href="state_estimation.html">Chapter 19: State Estimation</a></li>
<ul>
<li><a href=state_estimation.html#section1>Observers and the Kalman Filter</a></li>
<li><a href=state_estimation.html#section2>Recursive Bayesian Filters</a></li>
<li><a href=state_estimation.html#section3>Smoothing</a></li>
</ul>
<li><a href="rl_policy_search.html">Chapter 20: Model-Free Policy Search</a></li>
<ul>
<li><a href=rl_policy_search.html#section1>Policy Gradient Methods</a></li>
<ul>
<li>The Likelihood Ratio Method (aka
REINFORCE)</li>
<li>Sample efficiency</li>
<li>Stochastic Gradient Descent</li>
<li>The Weight Pertubation Algorithm</li>
<li>Weight Perturbation with an Estimated
Baseline</li>
<li>REINFORCE w/ additive Gaussian noise</li>
<li>Summary</li>
</ul>
<li><a href=rl_policy_search.html#section2>Sample performance via the signal-to-noise
ratio.</a></li>
<ul>
<li>Performance of Weight Perturbation</li>
</ul>
</ul>
<p style="margin-bottom: 0; text-decoration: underline;font-variant: small-caps;"><b>Appendix</b></p>
<li><a href="drake.html">Appendix A: Drake</a></li>
<ul>
<li><a href=drake.html#section1>Pydrake</a></li>
<li><a href=drake.html#section2>Online Jupyter Notebooks</a></li>
<li><a href=drake.html#section3>Running on your own machine</a></li>
<ul>
<li>Install Drake</li>
</ul>
<li><a href=drake.html#section4>Getting help</a></li>
</ul>
<li><a href="multibody.html">Appendix B: Multi-Body
Dynamics</a></li>
<ul>
<li><a href=multibody.html#section1>Deriving the equations of motion (an example)</a></li>
<li><a href=multibody.html#section2>The Manipulator Equations</a></li>
<ul>
<li>Recursive Dynamics Algorithms</li>
<li>Bilateral Position Constraints</li>
<li>Bilateral Velocity Constraints</li>
</ul>
<li><a href=multibody.html#section3>The Dynamics of Contact</a></li>
<ul>
<li>Compliant Contact Models</li>
<li>Rigid Contact with Event Detection</li>
<li>Time-stepping Approximations for Rigid Contact</li>
</ul>
<li><a href=multibody.html#section4>Parameter Estimation</a></li>
</ul>
<li><a href="optimization.html">Appendix C: Optimization and
Mathematical Programming</a></li>
<ul>
<li><a href=optimization.html#section1>Optimization software</a></li>
<li><a href=optimization.html#section2>General concepts</a></li>
<ul>
<li>Convex vs nonconvex optimization</li>
<li>Constrained optimization with Lagrange multipliers</li>
</ul>
<li><a href=optimization.html#section3>Convex optimization</a></li>
<ul>
<li>Linear Programs/Quadratic Programs/Second-Order Cones</li>
<li>Semidefinite Programming and Linear Matrix Inequalities</li>
<li>Sums-of-squares optimization</li>
<li>Solution techniques</li>
</ul>
<li><a href=optimization.html#nonlinear>Nonlinear programming</a></li>
<ul>
<li>Second-order methods (SQP /
Interior-Point)</li>
<li>First-order methods (SGD / ADMM) </li>
<li>Zero-order methods (CMA)</li>
<li>Example: Inverse Kinematics</li>
</ul>
<li><a href=optimization.html#section5>Combinatorial optimization</a></li>
<ul>
<li>Search, SAT, First order logic, SMT solvers, LP interpretation</li>
<li>Mixed-integer convex optimization</li>
</ul>
<li><a href=optimization.html#section6>"Black-box" optimization</a></li>
</ul>
<li><a href="playbook.html">Appendix D: An Optimization
Playbook</a></li>
</ul>
</section>
<section id="preface"><h1>Preface</h1>
<p>This book is about building robots that move with speed, efficiency, and
grace. I believe that this can only be achieve through a tight coupling
between mechanical design, passive dynamics, and nonlinear control synthesis.
Therefore, these notes contain selected material from dynamical systems
theory, as well as linear and nonlinear control.<p>
<p>These notes also reflect a deep belief in computational algorithms playing
an essential role in finding and optimizing solutions to complex dynamics and
control problems. Algorithms play an increasingly central role in modern
control theory; these days even rigorous mathematicians consider finding
convexity in a problem (therefore making it amenable to an efficient
computational solution) almost tantamount to an analytical result. Therefore,
the notes necessarily also cover selected material from optimization theory,
motion planning, and machine learning.</p>
<p> Although the material in the book comes from many sources, the
presentation is targeted very specifically at a handful of robotics problems.
Concepts are introduced only when and if they can help progress the
capabilities we are trying to develop. Many of the disciplines that I am
drawing from are traditionally very rigorous, to the point where the basic
ideas can be hard to penetrate for someone that is new to the field. I've made
a conscious effort in these notes to keep a very informal, conversational tone
even when introducing these rigorous topics, and to reference the most
powerful theorems but only to prove them when that proof would add particular
insights without distracting from the mainstream presentation. I hope that
the result is a broad but reasonably self-contained and readable manuscript
that will be of use to any enthusiastic roboticist.</p>
<section><h1>Organization</h1>
<p>The material in these notes is organized into a few main parts. "Model
Systems" introduces a series of increasingly complex dynamical systems and
overviews some of the relevant results from the literature for each system.
"Nonlinear Planning and Control" introduces quite general computational
algorithms for reasoning about those dynamical systems, with optimization
theory playing a central role. Many of these algorithms treat the dynamical
system as known and deterministic until the last chapters in this part which
introduce stochasticity and robustness. "Estimation and Learning" follows
this up with techniques from statistics and machine learning which
capitalize on this viewpoint to introduce additional algorithms which can
operate with less assumptions on knowing the model or having perfect
sensors. The book closes with an "Appendix" that provides slightly more
introduction (and references) for the main topics used in the course.</p>
<p>The order of the chapters was chosen to make the book valuable as a
reference. When teaching the course, however, I take a spiral trajectory
through the material, introducing robot dynamics and control problems one at
a time, and introducing only the techniques that are required to solve that
particular problem.</p>
<todo>insert figure showing progression of problems here. pendulum ->
cp/acro -> walking ... with chapter numbers associated.</todo>
</section>
<section><h1>Software</h1>
<p> All of the examples and algorithms in this book, plus many more, are now
available as a part of our open-source software project: <drake></drake>.
<drake></drake> is a C++ project, but in this text we will use Drake's <a
href="http://drake.mit.edu/python_bindings.html">Python bindings</a>. I
encourage super-users or readers who want to dig deeper to explore the C++
code as well (and to contribute back). </p>
<p>Please see the <a href="drake.html">appendix</a>
for specific instructions for using <drake></drake> along with these
notes.</p>
</section>
<p style="text-align:right;"><a href="intro.html">First chapter</a></p>
</section> <!-- end preface -->
<div id="footer">
<hr/>
<table style="width:100%;">
<tr><td><em>Underactuated Robotics</em></td><td align="right">© Russ
Tedrake, 2020</td></tr>
</table>
</div>
</body>
</html>
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手