Jacobian Analysis in Parallel Robots

The Jacobian matrix is the central tool for velocity analysis, force analysis, and singularity detection in robotic manipulators. For parallel robots, where multiple kinematic chains work together, the Jacobian reveals critical information about how the robot transmits motion and where it can fail.

Why the Jacobian Matters

The Jacobian tells us three things: (1) how fast the end-effector moves for given joint velocities, (2) how forces at the end-effector translate to joint torques, and (3) where singularities occur—configurations where the robot loses or gains degrees of freedom.

1. What is the Jacobian?

In calculus, the Jacobian is the matrix of all first-order partial derivatives of a vector-valued function. In robotics, it serves as the linear map between joint-space velocities and task-space velocities:

\dot{\mathbf{x}} = \mathbf{J}(\mathbf{q}) \, \dot{\mathbf{q}}

Where:

$\dot{\mathbf{x}} \in \mathbb{R}^m$ is the end-effector velocity (twist)
$\dot{\mathbf{q}} \in \mathbb{R}^n$ is the vector of joint velocities
$\mathbf{J}(\mathbf{q}) \in \mathbb{R}^{m \times n}$ is the Jacobian matrix

Figure 1. The Jacobian maps joint velocities to end-effector velocities.

The Dual Relationship: Forces

The Jacobian also connects end-effector wrenches to joint torques through the principle of virtual work:

\boldsymbol{\tau} = \mathbf{J}^T(\mathbf{q}) \, \mathcal{F}

Where $\boldsymbol{\tau}$ is the vector of joint torques and $\mathcal{F}$ is the wrench at the end-effector.

2. Serial vs. Parallel Jacobians

The structure of the Jacobian is fundamentally different for serial and parallel robots.

Aspect	Serial Robot	Parallel Robot
Velocity equation	$\dot{\mathbf{x}} = J\,\dot{\mathbf{q}}$	$\mathbf{A}\,\dot{\mathbf{x}} = \mathbf{B}\,\dot{\mathbf{q}}$
Jacobian form	$J$ directly	$J = \mathbf{A}^{-1}\mathbf{B}$
Singularity types	$\det(J) = 0$	$\det(\mathbf{A})=0$ or $\det(\mathbf{B})=0$
Inverse kinematics	Difficult (non-linear)	Easy (closed-form)
Forward kinematics	Easy (product of matrices)	Difficult (multiple solutions)

3. Deriving the Parallel Robot Jacobian

The general method follows three fundamental steps that work for any parallel mechanism.

Write the geometric constraint (loop closure)

For each kinematic chain, write the vector equation relating base, actuator variables, and platform pose: $\mathbf{f}(\mathbf{q}, \mathbf{x}) = \mathbf{0}$

Differentiate with respect to time

\frac{\partial \mathbf{f}}{\partial \mathbf{x}}\,\dot{\mathbf{x}} + \frac{\partial \mathbf{f}}{\partial \mathbf{q}}\,\dot{\mathbf{q}} = \mathbf{0}

Rearrange into standard form

Defining $\mathbf{A} = \frac{\partial \mathbf{f}}{\partial \mathbf{x}}$ and $\mathbf{B} = -\frac{\partial \mathbf{f}}{\partial \mathbf{q}}$:

\mathbf{A}\,\dot{\mathbf{x}} = \mathbf{B}\,\dot{\mathbf{q}} \qquad \Rightarrow \qquad \mathbf{J} = \mathbf{A}^{-1}\mathbf{B}

4. Complete Example: The Stewart Platform (6-UPS)

The Stewart–Gough platform is the most iconic parallel robot. Invented in 1965 for flight simulation, it is now used in manufacturing, surgery, astronomy, and motion simulators. It has 6 degrees of freedom (3 translations + 3 rotations), actuated by 6 linear actuators.

What does 6-UPS mean?

Each leg has: U (universal joint at the base) — P (prismatic actuator, actuated) — S (spherical joint at the platform). Six identical legs.

Interactive 3D Model

Drag to rotate · Scroll to zoom

Height: 0.40 m

Roll: 0.0°

Pitch: 0.0°

4.1 Geometric Constraint

The position of each platform joint $B_i$ in the fixed frame is:

B_i = \mathbf{p} + \mathbf{R}\, b_i

Where $\mathbf{p} = [x, y, z]^T$ is the platform center and $\mathbf{R}$ is the rotation matrix. The leg vector and length constraint:

\mathbf{d}_i = B_i - A_i \qquad \rho_i^2 = \mathbf{d}_i^T \mathbf{d}_i

4.2 Deriving the Velocity Equation

Differentiating and using the platform twist $\mathcal{V} = [\mathbf{v}^T, \boldsymbol{\omega}^T]^T$:

\dot{\rho}_i = \hat{u}_i^T \mathbf{v} + (\mathbf{R}\, b_i \times \hat{u}_i)^T \boldsymbol{\omega}

Where $\hat{u}_i = \mathbf{d}_i / \rho_i$ is the unit vector along leg $i$.

4.3 The Jacobian Matrix

Writing all 6 equations in matrix form:

\dot{\boldsymbol{\rho}} = \underbrace{\begin{bmatrix} \hat{u}_1^T & (\mathbf{R}\,b_1 \times \hat{u}_1)^T \\ \hat{u}_2^T & (\mathbf{R}\,b_2 \times \hat{u}_2)^T \\ \vdots & \vdots \\ \hat{u}_6^T & (\mathbf{R}\,b_6 \times \hat{u}_6)^T \end{bmatrix}}_{\mathbf{J}} \mathcal{V}

Stewart Platform Jacobian

Each row of the $6 \times 6$ Jacobian is the Plücker coordinates (reciprocal screw) of leg $i$. The relation $\dot{\boldsymbol{\rho}} = \mathbf{J}\,\mathcal{V}$ maps the platform twist to actuator velocities.

4.4 Numerical Example

Home Position

Platform at height $h = 0.4\,\text{m}$, no rotation ($\mathbf{R} = \mathbf{I}$). Base radius $R_b = 0.5\,\text{m}$, platform radius $R_p = 0.25\,\text{m}$.

For Leg 1:

$A_1 = [0.5, 0, 0]^T$, $b_1 = [0.217, 0.125, 0]^T$
$\mathbf{d}_1 = [-0.283, 0.125, 0.4]^T$, $\rho_1 = 0.509\,\text{m}$
$\hat{u}_1 = [-0.556, 0.246, 0.786]^T$

The first row of $\mathbf{J}$: $J_1 = [-0.556 \;\; 0.246 \;\; 0.786 \;\; 0.098 \;\; {-0.170} \;\; 0.035]$

4.5 Force Analysis

Using the Jacobian transpose for static force balance:

\mathbf{f} = \mathbf{J}^{-T} \mathcal{F}

Force Distribution Example

A downward load of $100\,\text{N}$ at the home position distributes to approximately $f_i \approx -21.2\,\text{N}$ per leg (compression). The symmetric geometry ensures all actuators share the load—a key advantage of parallel robots.

Screw Theory Interpretation

Each row of the Jacobian is the normalized wrench (zero-pitch screw) along leg $i$. This is the inverse Jacobian relationship, which is why inverse kinematics is straightforward for the Stewart platform.

5. Singularities

A singularity occurs when the Jacobian loses rank. For parallel robots, singularities are classified into three types.

Singularities Are Dangerous

At a singularity, the robot may:

Lose controllable degrees of freedom
Gain uncontrolled motions
Experience theoretically infinite actuator forces
Become structurally unstable

Type	Condition	Physical Meaning
Type I (Inverse)	$\det(\mathbf{B}) = 0$	Workspace boundary—actuator at its limit
Type II (Direct)	$\det(\mathbf{A}) = 0$	Platform moves with locked actuators
Type III (Combined)	Both $= 0$	Architecturally special geometry

For the Stewart platform, Type II singularities occur when the six leg lines belong to a linear complex—e.g., all intersecting a common line (Hunt singularity). The rows of the Jacobian are Plücker coordinates; their linear dependence is studied through Grassmann line geometry.

6. Physical Interpretation

Manipulability Index

w = \sqrt{\det(\mathbf{J}\,\mathbf{J}^T)} = \prod_{i=1}^{m} \sigma_i

When $w = 0$, the robot is at a singularity. The condition number $\kappa = \sigma_{\max}/\sigma_{\min}$ indicates proximity to singularity.

Stiffness

\mathbf{K} = \mathbf{J}^T \mathbf{K}_a \mathbf{J}

Platform stiffness depends on both actuator stiffnesses ($\mathbf{K}_a$) and the robot configuration through $\mathbf{J}$.

Summary

Key Takeaways

The Jacobian maps joint velocities to end-effector velocities
Parallel robots: $\mathbf{A}\dot{\mathbf{x}} = \mathbf{B}\dot{\mathbf{q}}$
Stewart platform Jacobian rows are Plücker coordinates of leg lines
Type I: workspace boundary · Type II: uncontrolled motion
Jacobian transpose relates forces: $\boldsymbol{\tau} = \mathbf{J}^T \mathcal{F}$
Manipulability $w = \sqrt{\det(\mathbf{J}\mathbf{J}^T)}$ — zero at singularities

Quick Check

Q1. Each row of the Stewart platform Jacobian represents:

A joint twist of an actuator

The Plücker coordinates of a leg line

The position vector of a base joint

Q2. A Type II singularity means:

The robot is at the workspace boundary

The platform can move with all actuators locked

The manipulability is maximized

Q3. The Stewart platform has how many DOF?

3 (position only)

4 (position + 1 rotation)

6 (full position + orientation)

← Previous: Screw Theory Next: Singularity-Free Workspace →