forked from RobotLocomotion/drake
-
Notifications
You must be signed in to change notification settings - Fork 0
/
ik.tex
100 lines (93 loc) · 6.83 KB
/
ik.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
\documentclass{article}
\usepackage{amsmath,listings,fullpage}
\title{Drake constraint and inverse kinematics}
\begin{document}
\maketitle
\section{Structure of Constraint}
Drake defines a base class \textsl{Constraint}, which has four sub classes: \textsl{SingleTimeKinematicConstraint}, \textsl{MultipleTimeKinematicConstraint}, \textsl{QuasiStaticConstraint} and \textsl{PostureConstraint}. For the mex version of Constraint object, calling \textsl{constraintTypemex} function would return the type string in MATLAB.
\subsection{SingleTimeKinematicConstraint}
This is for kinematic constraints being evaluated at a single time moment.
All \textsl{SingleTimeKinematicConstraint} objects share four methods: \text{getNumConstraint}, \textsl{eval}, \textsl{bounds} and \textsl{name}. For mex object, calling \textsl{testKinCnstmex} would return the number of constraints, the constraint value, bounds and name in MATLAB.
\begin{enumerate}
\item \textsl{WorldPositionConstraint} A box constraint on the position of some points on a certain body.
\item \textsl{WorldCoMConstraint} A box constraint on the position of the Center of Mass of the whole robot
\item \textsl{WorldQuatConstraint} Constraining the orientation of a given body by putting a constraint on the quaternion
\item \textsl{WorldEulerConstraint} A box constraint on the Euler angles of a given body, in the intrinsic z-y'-x'' order
\item \textsl{WorldGazeTargetConstraint} Constraining the orientation of a given body that a given axis of the body must gaze at a target. Support `cone threshold'
\item \textsl{WorldGazeDirConstraint} Constraining the orientation of a given body, that a given axis of the body must in a certain direction. Support `cone threshold'
\item \textsl{WorldGazeOrientConstraint} Constraining the orientation of a given body. Refer to the documentation in the implementation. Support `threshold' and `cone threshold'
\item \textsl{Point2PointDistanceConstraint} Constraining the distance between points on bodyA/world to points on bodyB/world to be within a range.
\item \textsl{WorldPositionInFrameConstraint} Constraining the point on bodyA in within a bounding box in a given frame.
\end{enumerate}
\subsection{MultipleTimeKinematicConstraint}
This is for kinematic constraint being evaluated at multiple time moments.
All \textsl{MultipleTimeKinematicConstraint} objects share four methods: \text{getNumConstraint}, \textsl{eval}, \textsl{bounds} and \textsl{name}. For mex object, calling \textsl{testKinCnstmex} would return the number of constraints, the constraint value, bounds and name in MATLAB.
\begin{enumerate}
\item \textsl{WorldFixedPositionConstraint} The position of certain points on a certain body is fixed for a given time interval.
\item \textsl{WorldFixedOrientConstraint} The orientation of a certain body is fixed for a given time interval.
\item \textsl{WorldFixedBodyPostureConstraint} The position and orientation of a certain body is fixed for a given time interval.
\end{enumerate}
\subsection{QuasiStaticConstraint}
Constraining that the \textit{Center of Mass} must lie within a shrunk support polygon. For mex object, call \textsl{testQuasiStaticConstraintmex} would return the active flag, the constraint value, bounds and names. The constraint is
\begin{equation}
c = \begin{bmatrix}x_{com}\\y_{com}\end{bmatrix}-w_1\begin{bmatrix}x_{pt1}\\y_{pt2}\end{bmatrix}-w_2\begin{bmatrix}x_{pt2}\\y_{pt2}\end{bmatrix}-\hdots-w_N\begin{bmatrix}x_{ptN}\\y_{ptN}\end{bmatrix}
\end{equation}
where $c$ is the constraint value. $x_{com},y_{com}$ is the x,y-position of the CoM, $w_i$ are the weights of associated with each shrunk point $pti$. Also, the user should examine the linear constraint
\begin{equation}
1 = w_1+w_2+\hdots+w_N
\end{equation}
This constraint is not included in the return value of \textsl{QuasiStaticConstraint.eval} since it is linear and easy to check.
\subsection{PostureConstraint}
This is a box constraint on the posture value. For mex object, calling \textsl{testPostureConstraintmex} would return the upper and lower bounds of all joints.
\subsection{Wrap Deprecated Constraint}
The method \textsl{wrapDeprecatedConstraint} would work to wrap the old format of constraint to \textsl{Constraint} object
\begin{lstlisting}[language=MATLAB]
[kc_cell,qsc_pts] = wrapDeprecatedConstraint(robot,body,pts,pos,options)
\end{lstlisting}
\section{Inverse Kinematics}
\subsection{inverseKin}
This method solves the inverse kinematics problem as
\begin{lstlisting}[language=MATLAB]
[q,info,infeasible_constraint] = ...
inverseKin(obj,q_seed,q_nom,constraint1,constraint2,constraint3,...,options)
\end{lstlisting}
Each constraint is a \textsl{Constraint} object. \textsl{options} is an \textsl{IKoptions} object. If the problem is infeasible, and the debug flag is on by calling
\begin{lstlisting}[language=MATLAB]
options = options.setDebug(true);
\end{lstlisting}
Then \textsl{infeasible\_constraint} would return the name of the constraints that are infeasible.
\subsection{inverseKinPointwise}
This method solves inverse kinematics problems sequentially for multiple time points
\begin{lstlisting}[language=MATLAB]
[q,info,infeasible_constraint] = ...
inverseKin(obj,t,q_seed,q_nom,kc1,kc2,kc3,...,options)
\end{lstlisting}
where $t$ is a row vector containing all the time moments to solve the inverse kinematics problem.
Setting \textsl{sequentialSeedFlag} to \textsl{true} in \textsl{ikoptions} would allow ik to use the solution at the previous time moment to be the seed of the current time moment.
\subsection{inverseKinTraj}
\begin{lstlisting}[language=MATLAB]
[xtraj,info,infeasible_constraint] = ...
inverseKinTraj(obj,t,q0,qd0,q_seed,q_nom,constraint1,constraint2,...,options)
\end{lstlisting}
This method solve the inverse kinematics problem for multiple time points simultaneously, and would output a smooth trajectory.
The problem it solves is
\begin{align}
\min_{q_1,q_2,...,q_N} \sum_{i = 1}^N &\ddot{q}_i'Q_a\ddot{q}_i+\sum_{i=2}^N \dot{q}_i'Q_v\dot{q}_i+\sum_{i=2}^N[q_i-q^{nom}_i]'Q[q_i-q^{nom}_i]\\
\text{subject to}&\quad \text{constraint } k\text{ at }t(j) \quad j = 1,2,\hdots N
\end{align}
The assumption is that $q$ is a cubic polynomial, with the initial value $q_0$ and initial time derivative $qd0$ at time $t(1)$.
Further samples can be added in between knot points to check if those samples satisfy the constraints. Check \textsl{setAddtionaltSamples} method in IKoptions.
\subsection{Wrap deprecated IK}
The method \textsl{inverseKinWrapup} would wrap the old IK input arguments to the new IK input arguments
\begin{lstlisting}[language=MATLAB]
[q_seed,q_nom,constraint,ikoptions] =
inverseKinWrapup(obj,q0,
body1,body1_pts,body1_pos,...
body1,body2_pts,body2_pos,...
body1,body3_pts,body3_pos,...
...
bodyN,bodyN_pts,bodyN_pos,...
options
use_mex_constraint)
\end{lstlisting}
\end{document}