Task-Space Singularity Avoidance for Control Affine Systems Using Control Barrier Functions

1. Problem Statement

The paper addresses the critical challenge of kinematic singularities in robotic and dynamical systems.

• The Issue: Singularities occur when the mapping from control inputs to task-space motion (the input-output matrix) loses rank. In these configurations, the system cannot generate forces or torques in certain directions, making it impossible to track arbitrary trajectories.
• Consequences: Near singularities, standard controllers relying on matrix inversion (e.g., Jacobian pseudoinverse) suffer from poor numerical conditioning. This leads to:
  • Large, abrupt spikes in control inputs (actuation).
  • Loss of stability and performance.
  • Safety risks, particularly in medical or safety-critical applications.
• Limitations of Prior Work: Existing methods (hybrid switching, hardware redesign, Lie-bracket synthesis, regularization) often lack formal safety guarantees, are difficult to generalize, or fail to provide real-time avoidance strategies for complex, state-dependent systems.

2. Methodology

The authors propose a Control Barrier Function (CBF) framework tailored for control-affine systems to ensure the system remains in a "safe" set where the input-output mapping retains full rank.

A. System Formulation

The system is modeled as a control-affine system:
$$\dot{x} = f(x) + g(x)u, \quad z = \gamma(x)$$
Where $x$ is the state, $u$ is the control input, and $z$ is the task-space output. The critical component is the state-dependent input-output mapping matrix:
$$\phi(x) = \frac{\partial \gamma}{\partial x}g(x) = J(x)g(x)$$
Singularities occur when $\phi(x)$ loses rank.

B. Singularity Identification via Eigenvalues

To detect singularities, the authors analyze the eigenvalues of a symmetric matrix derived from $\phi(x)$:

• For over-actuated systems ($d < m$), they use $M(x) = \phi(x)\phi(x)^T$.
• For other cases, they use $M(x) = \phi(x)^T\phi(x)$.
  The eigenvalues of $M(x)$ correspond to the squared singular values of $\phi(x)$. A singularity is approached when the smallest eigenvalue $\lambda_i(x)$ approaches zero.

C. CBF Construction

The authors define a safe set $C$ where the system maintains a safety margin $\epsilon$ from the singularity:
$$C_i = \{ x \mid \lambda_i(x) - \epsilon \geq 0 \}$$
The barrier function is defined as $h_i(x) = \lambda_i(x) - \epsilon$.
To ensure forward invariance (safety), the control input $u$ must satisfy the CBF condition:
$$\dot{h}_i(x) + \alpha(h_i(x)) \geq 0$$
This is enforced via a Quadratic Program (QP) that minimizes the deviation from a nominal control input $v(x)$ while satisfying the safety constraint.

D. Two Implementation Approaches

The paper presents two methods to construct these barrier functions:

1. Analytical CBF: Deriving explicit algebraic expressions for eigenvalues and their derivatives. This is feasible for simple systems (e.g., 2-link manipulators) but becomes algebraically intractable for complex systems.
2. Numerical CBF (Obstacle Avoidance Reformulation): For complex systems (e.g., magnetic actuation), where analytical derivatives are too complex:
   • The unsafe region (near-zero singular values) is identified numerically using Singular Value Decomposition (SVD) to generate a point cloud of singular configurations.
   • This point cloud is converted into a 3D "virtual obstacle" using Alpha Shapes to create a triangulated boundary.
   • The problem is reformulated as a standard obstacle avoidance task, where the distance to the nearest point on the obstacle boundary serves as the barrier function. This avoids the need for complex symbolic differentiation.

3. Key Contributions

• Unified Framework: A theoretical framework applicable to underactuated, fully actuated, and over-actuated control-affine systems.
• Theoretical Guarantees: Proof that the proposed CBF-QP renders the safe set forward invariant, ensuring the system never enters a singular configuration (provided actuator constraints allow).
• Analytical and Numerical Solutions: The paper bridges the gap between theoretical derivation and practical application by offering an analytical method for simple systems and a robust numerical "obstacle avoidance" method for complex, high-dimensional systems.
• Regularized QP: Introduction of a specific cost function formulation that handles rank-deficient cost matrices in over-actuated systems by adding a regularization term $\Gamma$.

4. Results

The framework was validated through simulations on two distinct systems:

A. Planar 2-Link Manipulator

• Scenario: The manipulator was tasked with moving toward a singular configuration (arm fully straightened, joint angle $q_2 \to 0$).
• Outcome:
  • Without CBF: The controller attempted to straighten the arm, causing joint velocities to spike by 70× to 100× due to Jacobian ill-conditioning.
  • With CBF: The controller modified the joint angles to maintain a safety margin, resulting in smooth trajectories and eliminating control spikes.

B. Magnetically Actuated Needle (Medical Suturing)

• Scenario: A needle controlled by 4 electromagnets performing a suturing task. The system is over-actuated and highly non-linear.
• Outcome:
  • The numerical CBF approach successfully identified singular regions in the 3D configuration space (position + orientation).
  • The needle followed the desired path while slightly tilting its orientation to avoid singularities.
  • Performance: The system achieved an RMS position error of 0.99 mm and orientation error of 0.65 rad, with a maximum current of 4 A. The needle successfully avoided the singular set (depicted in green in the configuration space plot) while completing the stitch.

5. Significance

This work provides a rigorous, real-time solution to a long-standing problem in robotics: singularity avoidance with formal safety guarantees.

• Safety-Critical Applications: The method is particularly vital for medical robotics (e.g., magnetic needle steering) and autonomous systems where failure due to singularities could be catastrophic.
• Scalability: By introducing the numerical "obstacle avoidance" approach, the authors overcome the algebraic complexity that previously limited CBF application to simple systems.
• Control Quality: The method significantly reduces control input spikes, leading to smoother, more stable, and energy-efficient operation compared to nominal controllers.

In summary, the paper transforms singularity avoidance from a heuristic or optimization-based problem into a mathematically guaranteed safety constraint, enabling robust control of complex dynamical systems in the presence of rank-deficient configurations.