# Task-Space Motion Model

The task-space motion model characterizes the motion of a manipulator under closed-loop task-space position control, as used in the `taskSpaceMotionModel`

object and Task Space Motion Model block.

Robot manipulators are typically position controlled devices. For task-space control, you specify a reference end-effector pose in SE(3), and the model returns the joint configuration vector $$q$$ and its state derivatives. You can do this by using closed-loop control on the robot joints and using the motion model to simulate the behavior of the robot under this control.

For this approach to most closely approximate the motion of the actual system, you must accurately represent the dynamics of the controller and plant. While there are many variations of task-space control, this model uses a relatively simple Jacobian-Transpose approach that is most accurate when the dynamic parameters of the robot and controller are well selected to match the deisred behavior. To do this, you must understand the key parameters and motion model behavior:

### Background

#### Joint-Space vs. Task-Space Motion Models

There are two categories of robot position control:

Joint-Space motion control — In this case, the position input of the robot is specified as a vector of joint angles or positions, known as the robot's

*joint configuration,*$$q$$. The controller tracks a reference configuration, $${q}_{ref}$$, and returns the actual joint configuration $$q$$. This is also known as*configuration-space*control.Task-Space

*operational-space*control.

These figures shows the different types of inputs/outputs in these two motion control categories.

This topic page is specific to task-space motion control, as used in the `taskSpaceMotionModel`

object and Joint Space Motion Model block. For joint-space motion models, see the `jointSpaceMotionModel`

object. For an example that covers the difference between task-space and joint-space control in greater detail, see Plan and Execute Task- and Joint-Space Trajectories Using KINOVA Gen3 Manipulator.

#### Usage in MATLAB® and Simulink®

The task-space motion model can be represented in MATLAB or Simulink.

In Simulink, the Task-Space Motion Model block accepts the reference inputs and optional external force when applicable, and returns the joint configuration, velocity, and acceleration. The block handles integration, so no additional integration is required.

In MATLAB, the

`taskSpaceMotionModel`

system object models the closed-loop motion. The`derivative`

method returns derivatives of the joint configuration, velocity, and acceleration at each instant in time, so you must use an ODE solver or equivalent external integration method to simulate the motion in time.

For a more specific overview, refer to the associated documentation pages.

### Key Variables

The model state consists of these values:

$$q$$ — Robot joint configuration, as a vector of joint positions. Specified in $$rad$$ for revolute joints and $$m$$ for prismatic joints.

$$\underset{}{\overset{\dot{}}{q}}$$ — Vector of joint velocities in $$rad\cdot {s}^{-1}$$ for revolute joints and $$m\cdot {s}^{-1}$$ for prismatic joints

$$\underset{}{\overset{\xa8}{q}}$$ — Vector of joint accelerations in $$rad\cdot {s}^{-2}$$ for revolute joints or $$m\cdot {s}^{-2}$$ for prismatic joints

The end-effector pose $$T(q)$$ of the robot is a 4-by-4 homogeneous matrix defined relative to the origin at the robot base. Positions are in meters. Two forms of $$T$$ are used for calculating errors in control:

$${T}_{ref}$$ — Reference end-effector pose, specified as a desired end-effector pose

$${T}_{act}$$ — Actual end-effector pose achieved by the motion

The end-effector transform decomposes as:

$$T=\left[\begin{array}{cc}R& X\\ 0& 1\end{array}\right]$$

where $$R$$ is the orientation as a 3-by-3 rotation matrix, and $$X$$ is a 3-by-1 vector of *xyz*-positions in meters.

The task-space velocity $$v$$ and acceleration $$a$$ consist of two 6-by-1 vectors:

$$v=\left[\begin{array}{c}\omega \\ \underset{}{\overset{\dot{}}{X}}\end{array}\right]$$, $$a=\left[\begin{array}{c}\alpha \\ \underset{}{\overset{\xa8}{X}}\end{array}\right]$$

where $$\omega $$ and $$\alpha $$ are 3-by-1 vectors of angular velocities and accelerations of the frame, respectively.

### Equations of Motion

Use the task-space motion model to represent robots that are subject to a control law that acts on the task-space error. For example, when the input to the control law specifies end-effector motion. While there are many ways to implement such a system, in this model, the closed-loop response is approximated in the Plan and Execute Task- and Joint-Space Trajectories Using KINOVA Gen3 Manipulator object by providing a system under proportional-derivative Jacobian-Transpose style control. Regardless of how the system you are using works, this model can be used as a low-fidelity approximation of a system under closed-loop task-space control.

#### Proportional-Derivative Control

When the motion model uses proportional-derivative (PD) control, as determined by the MotionType property of the `taskSpaceMotionModel`

object, the model computes forward dynamics using standard rigid body dynamics, but with subject to a PD control law that acts on the error between desired and actual end-effector pose.

Inputs —

Outputs —

Complexity —

When it can be applied —

In this system, the joint positions, velocities, and accelerations are computed using standard rigid-body robot dynamics. For more information, see Robot Dynamics. The generalized force input $$Q$$ is given by the PD Control law on the task-space error, scaled to the joint-space using Jacobian-Transpose style control:

$$\frac{d}{dt}\left[\begin{array}{c}q\\ \underset{}{\overset{\dot{}}{q}}\end{array}\right]={f}_{dyn}(q,\underset{}{\overset{\dot{}}{q}},Q,{F}_{ext})$$

$$Q=J(q{)}^{T}({K}_{P}{E}_{T}+{K}_{D}{E}_{v})-B\underset{}{\overset{\dot{}}{q}}+G(q)$$

$${E}_{T}=\left[\begin{array}{c}{e}_{rot}\\ {e}_{trans}\end{array}\right]$$

$${E}_{v}=({v}_{ref}-J(q)\underset{}{\overset{\dot{}}{q}})$$

where:

$${e}_{rot}$$ — is the rotational error converted to Euler angles using

`rotm2eul(`

$${R}_{ref}{R}_{act}^{T}$$).$${e}_{pos}$$ — is the error in

*xyz*-coordinates, calculated as $${X}_{ref}-{X}_{act}$$.$$G(q)$$ — are the gravity torques and forces for all joints to maintain their positions in the specified gravity. For more information, see the

`gravityTorque`

function.$$J(q)$$ — is the geometric Jacobian for the given joint configuration for more information, see the

`geometricJacobian`

function.

The control input relies on these user-defined parameters:

$${K}_{P}$$ — Proportional gain, specified as a 6-by-6 matrix

$${K}_{D}$$ — Derivative gain, specified as a 6-by-6 matrix

$$B$$ — Joint damping vector, specified as a two-element vector of damping constants in $$N\cdot s\cdot ra{d}^{-1}$$ for revolute joints and $$N\cdot s\cdot {m}^{-1}$$ for prismatic joints

You can specify these parameters as properties on the `taskSpaceMotionModel`

object.

This model accepts the following inputs:

$${T}_{ref}$$ — Reference end-effector pose, specified as the desired end-effector pose

$${v}_{ref}$$ — Reference end-effector velocities, specified as a vector $$v=\left[\begin{array}{c}\omega \\ \underset{}{\overset{\dot{}}{X}}\end{array}\right]$$, with angular velocities $$\omega $$ and translational velocities $$\underset{}{\overset{\dot{}}{X}}$$