rigidbody.texp

\chapter{Rigid Body Systems} \label{c:rigidbody}

Overview of specific methods for kinematic chains.

\section{Forward Kinematics}

\section{Inverse Kinematics}

\section{Kinematic Planners}

\section{Contact Dynamics}


\section{Aerodynamics}

Drake includes the ability to incorporate aerodynamic forces into a
model through the use of the RigidBodyWing class, and is
flexible on how these forces are modeled. The programs AVL and Xfoil,
written in part by Mark Drela are used to determine proper lift, drag,
and moment coefficients for specified 3D airfoils on rectangular
wings. The Wing class actually supports four ways of defining an
airfoil for the model:
\begin{itemize}
\item Empirical data for lift, drag, and moment coefficients. These should be in a .mat file with
variable names ``CLSpline, CDSpline, CMSpline''
\item A 4 or 5-digit NACA airfoil code. i.e. ``NACA0012''
\item The words ``flat plate''
\item A .dat file (Xfoil output) specifying x-y points of the surface of the airfoil. See Xfoil's
documentation for more information. This file should be located somewhere in the Matlab
path.
\end{itemize}
The properties of the airfoil such as the chord, span, stall angle,
and nominal speed are user inputted. The Wing class will display a message during initialization if it detects the wing stalls before the angle
the user inputted angle.

\subsection{Post-stall points}

Before the user-specified stall angle, the lift, drag, and moment
coefficients for .dat files and NACA airfoils are calculated by AVL
and Xfoil. After the user-specified stall angle, force and moment
coefficients are still included all the way to $\pm 180$ degrees angle of
attack. These are calculated from flat plate theory (which is
supported by experimental data taken by Joseph Moore), so there are
some inaccuracies due to camber and thickness of the actual airfoil,
and the fact that post-stall is extremely difficult to model without
taking data in a wind tunnel. However, highly dynamic knife-edge
experiments using a ``wingeron'' aircraft support using this technique
as a reasonable starting point in a model such that trajectories and
controllers can be developed. Further improvements to the model via
system identification on actual data can also be made.

\subsection{Modeling flat plates}

The Wing class makes an improvement to the flat plate model that Rick
used for perching which deals with the point of application of lift
and drag forces on the wing (pitching moment coefficient). The
aerodynamic center of a flat plate airfoil pre-stall is at the
quarter-chord\footnote{\url{https://courses.cit.cornell.edu/mae3050/mae3050ThinAirfoils.pdf}}, meaning that pre-stall, there is no pitching moment
about the quarter-chord point on the plate. However, intuition and
research\footnote{Stall flutter and nonlinear divergence of a
  two-dimensional flat plate wing / John Dugundji, Krishnaswamy Aravamudan. TL570.M41.A25 no.159-6} suggests that when a plate is oriented perpendicular to
oncoming flow, the zero-moment point on the plate is in the
center. This would imply a moment about the quarter-chord point. For
any wing, the center of pressure is modeled as moving rearward from
the quarter-chord to the middle of the plate starting at stall and
ending at 90 degrees angle of attack.

\subsection{Generating the Lift, Drag, and Moment splines}

Template files for inputs to AVL and xfoil are provided in the
RigidBodyWing folder. These have properties starting with the \$ tag
that are replaced by appropriate values calculated from the parameters
of the Wing constructor, which originate from the tags in the model's
URDF. If for some reason Xfoil and AVL cannot run correctly (the most
likely cause being they cannot find a proper input file), then a
plaintext warning message will print, but the splines will still be
constructed. If AVL fails, then the splines.

Both Planar and full 3-D classes of Wing are supported, and these differ almost exclusively in their
computeSpatialForce method. The construction of the Wing object is
virtually exactly the same.

\subsection{Other things to note}

\begin{itemize}
\item The direction of rotations between standard aerodynamic
  convention and the coordinate frame used for wings: For a coordinate
  frame that is directed with X=forwards, Y=out the left of a wing,
  and Z=up, a positive-Y rotation will be a pitch downward. However,
  this is different than standard aerodynamic convention (which XFOIL
  and AVL outputs use), which define a pitch up as a positive
  rotation. This XYZ coordinate frame that the Wing class expects.
\item RPY support for the wing origin is purposefully excluded. The wing must have the same
coordinate axes as its parent body (but not the same xyz origin). If you need rotations, define
the parent body with the proper rotations.
\item The generated splines are dimensionalized. Lift (and Drag) force
  is equal to: $L=.5⋅Cl (\alpha)\rho v^2 S$ with $S$=wing area,
  $\rho$=air density, $v$=velocity. That is, the splines include the
  $.5\rho S$. The moment spline also included the chord length term
  present in the moment equation. Evaluating a spline at a given angle
  of attack ($\alpha$), and multiplying by $v^2$ will give the
  appropriate force or torque.
\end{itemize}

\subsection{Thrust components}

See the Drake documentation for details on how Thrust elements are
defined in a URDF. Thrust components have an xyz orientation and XYZ
direction which define the point and orientation of the applied
force. These vectors should be normalized, and the scaleFactor should
be used to translate the input command into Newtons of force. The
Thrust class works by returning a \mcode{B_modifications} matrix that is added
to \mcode{B} to properly capture the input force's effects on the dynamics of
the robot.

\begin{comment}
\subsection{ComputeSpatialForce}

computeSpatialForce returns $f_{ext}$ (a potentially sparse matrix with
manip.getNumBodies columns) and $B$ (a (nq x nu) matrix which
contributes a control-affine term + $B(q,qd)u$ to the manipulator
dynamics). Typically one of these will be a matrix of zeros while the
other will be nonzero, depending on whether the force is derived from
the state (springs, dashpots, wings) or the force is derived from an
input (Thrust). The direct_feedthrough_flag can indicate which is the
case, and is true for the latter.  Check out manipulatorDynamics to
see exactly how the $f_{ext}$ and $B_{mod}$ are included to the manipulator
equations. $F_{ext}$ gets passed through to the Featherstone model (after
a re-mapping to ensure the columns in $f_{ext}$ are in the proper order
that the Featherstone engine expects), while the $B_{mod}$ simply gets
added to the $B$ matrix.  The comments in Bug 1649, and the section
below can also be useful in understanding Thrust elements.


\subsection{Updating Plane.URDF}

Whenever the plane model is updated significantly due to moving around
weight, increasing the size of wings, or anything else that affects
the dynamics or collision geometry, the URDF should be updated.  Some
simple changes could be made manually if you know exactly how the
model changed, but anything that changes the inertial properties
should probably go through the Solidworks Export process again: 
\begin{enumerate}
\item Go
through the steps to export the URDF from the Solidworks model in the
Drake documentation.  
\item Put the meshes generated in the appropriate
location, and make sure the <visual> tags all reference the correct
.obj meshes. Copy the material colors from the old URDF to the new
URDF if desired.  
\item Copy all the collision tags from the old model
into the new URDF. Update these as necessary if the geometry of the
plane changed.  
\item Copy all the <force_element> tags from the old
model to the new model. If the geometry of the plane didn't change,
you shouldn't have to change anything in these tags. If the chord,
span, etc of any of the wings or surfaces changed, then update the
corresponding force tag as necessary.  
\end{itemize}
Example: You move the motor 4
cm forwards on the plane, and update the SolidWorks accordingly.  You
export the new URDF from the updated model, and then make the
following changes to the generated URDF: 
\begin{itemize}
\item Update mesh location for
visual tag (and convert meshes to .obj files if necessary) 
\item Copy all
collision properties from old plane. Update the Fuselage collision box
because the plane is now 4 cm longer (make the box 4 cm longer (x
dimension), and move the collision reference origin. This collision
reference origin may or may not move by 2 cm, depending on how the
fuselage moved relative to the defined model origin when you exported
from SolidWorks.) The collision tags for the other links should not
have to be modified. 
\item Copy all the force elements from the old
plane. The wing elements should not have to change, but the xyz origin
of the Propellers thrust element should have to be moved forward.
\end{itemize}
When in doubt, if you're not sure where to place collision or
force_element reference origins, set the <visual> tag to Origin:<xyz =
“0 0 0”/>, Geometry:<sphere radius =“.005”/>, and observe where the
link's origin is. Then grab a ruler, the actual airplane, and figure
out where to center the box such that it is in the location of the
corresponding surface. Modify the box in the <visual> tag so you can
see what is happening, and when you're happy, switch it over to the
<collision> tag, and reset the <visual> tag to Origin: <xyz = “0 0
0”/> and Geometry pointing to the appropriate mesh.


\subsection{Growth of state space from using URDFs}

Many of the aircraft originally simulated employ position control for
their control surfaces—This is currently not supported from a
URDF. This means that any actuated surfaces will add 2 extra DOF to
the model as well as the input required for the
actuator. Velocity-controlled joints should add 1 extra DOF on top of
the input required. With the inertial properties known, the proper
torque can be computed for a desired input position or velocity. Or,
the expanded-state model can be used in simulation, as position
control of the joints on the actual aircraft is effectively the
expanded-state model with high-gain controllers on top of some states
to drive them to the commanded values very quickly.

\subsection{Debugging}

If you need to debug the Wing class, it is recommended to step through
the system commands that the constructor generates, manually run
these, and check their output. There can be some quirks (such as Xfoil
cannot handle filenames longer than 64 characters) that might be
otherwise hard to track down. Also, check the input and output files
that get generated in Matlab's temporary directory to make sure they
look okay.

For forces and moments, keep in mind that Xfoil/AVL treat a positive
pitch up, while drake will usually treat a positive pitch as a pitch
down for a standard X-forward, Z-up axis configuration. Check the
jsign property in Planar robots, though.

\subsection{References and Useful Papers}

C.C. Critzos, H.H. Heyson, R.W. Boswinkle Jr. “Aerodynamic characteristics of NACA 0012 airfoil at
angles of attack from 0 to 180 degrees”. NACA TN 3361, 1955

R.E. Sheldahl, P.C. Klimas. “Aerodynamic Characteristics of Seven Symmetrical Airfoil Sections
Through 180-Degree Angle of Attack for Use in Aerodynamic Analysis of
Vertical Axis Wind Turbines. Report SAND80-2114, Sandia Laboratories, Albuquerque, March 1981.

Stall flutter and nonlinear divergence of a two-dimensional flat plate wing / John Dugundji,
Krishnaswamy Aravamudan. TL570.M41.A25 no.159-6

\end{comment}