5.3. Singularities and Manipulability

Singularities and Manipulability

A configuration ${\Theta }^{\ast }$ is a singularity of a manipulator if the Jacobian loses rank there:
$\mathrm{rank}(J({\Theta }^{\ast }))<\min (6,n).$
At a singularity, there exist task-space directions in which the end-effector cannot move regardless of joint velocities, and the inverse mapping from task-space velocities to joint velocities becomes ill-defined or unbounded. Singularities partition the configuration space $\mathcal{Q}$ into regular (full-rank) and singular subsets. Generically, the singular set has measure zero in $\mathcal{Q}$—it forms lower-dimensional surfaces, not open regions—but it can pose serious practical problems when a trajectory passes near or through it.

Singularities are classified by their geometric origin:

• Shoulder singularity: the wrist center lies on the axis of joint 1 (the base joint), making the base rotation redundant with respect to positioning.
• Elbow singularity: two or more links become collinear, so the arm is fully extended or folded. The manipulator loses radial reach capability.
• Wrist singularity: two wrist joint axes align, collapsing two rotational degrees of freedom into one (analogous to gimbal lock in Euler angles).

Manipulability quantifies how “far” a configuration is from singularity and how effectively the manipulator can generate velocities in all task-space directions. Three common scalar measures, all derived from the singular values ${\sigma }_1\ge \cdots \ge {\sigma }_r$ of $J(\Theta )$, are:
${\mu }_1={\sigma }_{\min }(J),{\mu }_2=\frac{{\sigma }_{\min }}{{\sigma }_{\max }},{\mu }_3=\sqrt{\det (J{J}^T)}.$

Near-Singular Behavior ${\sigma }_{\min }\to 0$, so the pseudoinverse $J^{†}$ amplifies noise and small task-space commands produce enormous joint velocities. This is why damped least squares is preferred over the raw pseudoinverse near singular configurations.

Near a singularity,

Intuition

The Jacobian maps the unit ball in joint-velocity space to an ellipsoid in task-space velocity. The principal axes of this ellipsoid are the left singular vectors of $J$, and their lengths are the singular values ${\sigma }_i$. At a singularity, one or more singular values collapse to zero: the ellipsoid degenerates, losing a dimension, and the manipulator cannot produce velocity in the corresponding direction.

Manipulability measures capture the “shape” of this ellipsoid. The minimum singular value ${\mu }_1={\sigma }_{\min }$ measures the worst-case velocity capability—the thinnest direction of the ellipsoid. The condition number ${\mu }_2={\sigma }_{\min }/{\sigma }_{\max }$ measures how isotropic the ellipsoid is: ${\mu }_2=1$ means the manipulator can move equally well in all directions (a sphere), while ${\mu }_2\to 0$ means it can barely move in one direction relative to another. Yoshikawa’s measure ${\mu }_3=\sqrt{\det (J{J}^T)}={\prod }_i{\sigma }_i$ is proportional to the volume of the ellipsoid—it captures overall velocity capability in all directions simultaneously.

The classification into shoulder, elbow, and wrist singularities corresponds to which part of the kinematic chain causes the rank deficiency. For a 6R manipulator with a spherical wrist (joints 4–6 intersecting), the Jacobian approximately decouples into a $3\times 3$ positioning block (joints 1–3) and a $3\times 3$ orientation block (joints 4–6), and singularities arise when either block loses rank independently.

Properties

Singular Value Decomposition of the Jacobian

(SVD) Any Jacobian $J\in {\mathbb{R}}^{6\times n}$ admits a decomposition $J=U\Sigma V^T$, where $U\in {\mathbb{R}}^{6\times 6}$ and $V\in {\mathbb{R}}^{n\times n}$ are orthogonal, and $\Sigma \in {\mathbb{R}}^{6\times n}$ is diagonal with entries ${\sigma }_1\ge {\sigma }_2\ge \cdots \ge {\sigma }_r\ge 0$ where $r=\min (6,n)$. The columns of $U$ (left singular vectors) are the principal directions of the velocity ellipsoid in task space; the columns of $V$ (right singular vectors) are the corresponding directions in joint space.

(Velocity Ellipsoid) The image of the unit ball $\Vert \dot{\Theta }\Vert \le 1$ under $J$ is the ellipsoid $\{V\in {\mathbb{R}}^6:{\sum }_i(u_i^{T}V/{\sigma }_i{)}^2\le 1\}$ with semi-axes ${\sigma }_i{u}_i$. At a singularity, one or more semi-axes vanish and the ellipsoid collapses.

Singularity Classification

(Formal Condition) ${\Theta }^{\ast }$ is singular if and only if $\det (J({\Theta }^{\ast })J({\Theta }^{\ast }{)}^T)=0$ (for $n\ge 6$) or $\det (J({\Theta }^{\ast }{)}^{T}J({\Theta }^{\ast }))=0$ (for $n\le 6$). Equivalently, the rows (or columns) of $J$ become linearly dependent.

(Shoulder Singularity) Occurs when the wrist center $p_w$ lies on the $z$-axis of the base joint ($z_0$). Joint 1 rotation then moves the wrist center in a circle of zero radius—it cannot change $p_w$ at all. The first column of $J$ becomes linearly dependent on other columns.

(Elbow Singularity) Occurs when the arm is fully extended (${\theta }_3=0$ for a typical elbow joint) or fully folded (${\theta }_3=\pi$). The two links become collinear, and the manipulator cannot move the wrist center radially (toward or away from the shoulder).

(Wrist Singularity) Occurs when two of the three wrist axes align. For a ZYZ Euler wrist (joints 4–6), this happens when ${\theta }_5=0$ or ${\theta }_5=\pi$, making joints 4 and 6 rotate about the same axis. This is identical in mechanism to gimbal lock.

(Measure Zero) The singular set $\{\Theta :\det (J{J}^T)=0\}$ is the zero set of an analytic function on $\mathcal{Q}$. For a non-degenerate manipulator, this is a set of measure zero—singularities form surfaces (or lower-dimensional manifolds) in joint space, not open regions. Almost every configuration is regular.

Manipulability Measures

(Minimum Singular Value) ${\mu }_1(\Theta )={\sigma }_{\min }(J(\Theta ))$. This is the most conservative measure: it equals the minimum task-space velocity achievable with a unit joint-space velocity. At a singularity, ${\mu }_1=0$. Disadvantage: it ignores the overall velocity capability and focuses only on the worst direction.

(Condition Number) ${\mu }_2(\Theta )={\sigma }_{\min }/{\sigma }_{\max }\in [0,1]$. Measures the isotropy of velocity generation. ${\mu }_2=1$ is the ideal isotropic configuration where the manipulator moves equally well in all directions. For a $6\times 6$ Jacobian, $\kappa (J)=1/{\mu }_2$ is the matrix condition number.

(Yoshikawa’s Measure) ${\mu }_3(\Theta )=\sqrt{\det (J{J}^T)}={\prod }_{i=1}^{\min (6,n)}{\sigma }_i$. Proportional to the volume of the velocity ellipsoid. Vanishes at singularities and is maximized at configurations where the manipulator has the greatest overall velocity capability. This is the most commonly used scalar manipulability index.

(Maximizing Manipulability) In redundant manipulators ($n>6$), the null-space freedom can be used to maximize a manipulability measure while tracking a desired end-effector trajectory. The gradient ${\nabla }_{\Theta }{\mu }_3$ can be projected into the null space of $J$ to improve manipulability without affecting the task (see null-space optimization).

Examples

2R Planar Arm Singularity

For a 2R arm with link lengths ${\ell }_1={\ell }_2=1$, the planar Jacobian (mapping $({\dot{\theta }}_1,{\dot{\theta }}_2)$ to $(\dot{x},\dot{y})$) is:

Singularity analysis $J_p=\left[\begin{array}{cc}-{\ell }_1{s}_1-{\ell }_2{s}_{12}& -{\ell }_2{s}_{12}\\ {\ell }_1{c}_1+{\ell }_2{c}_{12}& {\ell }_2{c}_{12}\end{array}\right]$ where $s_{12}=\sin ({\theta }_1+{\theta }_2)$, $c_{12}=\cos ({\theta }_1+{\theta }_2)$.

$\det (J_p)={\ell }_1{\ell }_2\sin ({\theta }_2)$.

Singularity at ${\theta }_2=0$: arm fully extended, length ${\ell }_1+{\ell }_2=2$. The end-effector can only move tangentially (perpendicular to the arm), not radially. Both columns of $J_p$ become parallel.

Singularity at ${\theta }_2=\pi$: arm folded back on itself. The end-effector is at distance $|{\ell }_1-{\ell }_2|=0$ from the base (for equal links), and again only tangential motion is possible.

Manipulability: ${\mu }_3=|\det (J_p)|=|\sin {\theta }_2|$. Maximum at ${\theta }_2=\pm \pi /2$ (elbow bent at right angle), zero at full extension/folding.

Wrist Singularity of a 6R Arm

A spherical wrist with ZYZ Euler angles $({\theta }_4,{\theta }_5,{\theta }_6)$ has orientation Jacobian proportional to:

Gimbal lock mechanism $J_{\text{wrist}}\propto \left[\begin{array}{ccc}0& -s_4& c_4{s}_5\\ 0& c_4& s_4{s}_5\\ 1& 0& c_5\end{array}\right]$

$\det (J_{\text{wrist}})\propto s_5$. When ${\theta }_5=0$, joints 4 and 6 both rotate about the same axis ($z$-axis of the wrist), so ${\dot{\theta }}_4$ and ${\dot{\theta }}_6$ produce identical end-effector motion. The rank drops from 3 to 2. Near ${\theta }_5=0$, achieving a desired angular velocity about a direction perpendicular to the aligned axis requires ${\dot{\theta }}_4$ or ${\dot{\theta }}_6$ to approach infinity.