Admittance Control of a Multi-degree-of-freedom Robot using an Asymmetric Matrix

Toshiaki Tsuji, Yasuhiro KatoJournal of the Robotics Society of Japan 2024 · DOI

Abstract

In contact-rich tasks, robots can correct positional errors by generating compliant motion. Methods using compliance control have the advantage of easily adjusting parameters. However, employing an asymmetric matrix as the stiffness matrix extends design freedom while posing a challenge in demonstrating passivity. This study includes a stability analysis in 3-dimensional space based on the root locus method and proposes a parameter design method based on the analysis. The asymmetric part generates a curl force field that may cause instability. Nonetheless, by designing the asymmetric part such that the eigenvalues of the stiffness matrix exclude imaginary components, vibration resulting from the rotational force field can be avoided.

Introduction

In robotics, compliance control allows a robot to react to external forces, which is essential for tasks like inserting a peg into a hole or collaborating with humans. Traditionally, engineers use symmetric, positive-definite matrices for the robot's stiffness. This ensures "passivity," meaning the robot doesn't "create" energy and remains inherently stable.

However, symmetric matrices are limiting. If you want a force in the $Z$ direction to specifically induce a movement in the $X$ direction (to "steer" the robot into a hole, similar to a mechanical Remote Center Compliance device), you need off-diagonal terms in your stiffness matrix. Using an asymmetric matrix provides this flexibility but introduces a major problem: it creates a non-conservative "curl" force field. This field can pump energy into the system, causing vibrations or total instability.

Comparison of a symmetric convergent force field (left) and an asymmetric curl force field (right) that can cause instability.

Methodology

The authors move away from the restrictive passivity framework and instead use classical control theory tools to analyze stability in 3D space. The admittance model (mass-damping-stiffness) is defined as a state-space system, and the system's poles are analyzed via the root locus method as damping parameters change.

Block diagram of the admittance control architecture, including the position controller and disturbance observer (DOB).

The Triangular Matrix Strategy

The key insight is that vibrations occur when the stiffness matrix has imaginary components in its eigenvalues. To avoid this, the authors propose using triangular matrices. Mathematically, the eigenvalues of a triangular matrix are simply its diagonal elements. This allows the inclusion of off-diagonal "steering" terms while keeping the stability analysis as simple as a standard diagonal matrix. This approach effectively creates a "software-defined" version of the Remote Center Compliance (RCC) mechanism, allowing the steering direction and stiffness to be changed instantly via code.

Key Results

The system was evaluated through numerical simulations and physical experiments on a 6-axis manipulator.

  • Stability Verification: In simulations, a general asymmetric matrix with imaginary eigenvalues caused the robot to vibrate. In contrast, the proposed triangular matrix allowed the robot to converge smoothly to its equilibrium point without oscillation.
  • Experimental Success: Using a 6-axis manipulator, the authors showed that a force applied in the $Z$-axis (pushing against a surface) could successfully and stably induce movements in the $X$ and $Y$ axes.
  • Performance: Previous research using similar asymmetric designs reduced peg-in-hole insertion times to less than half of standard methods. This paper provides the theoretical "safety net" to make those high-performance designs reliable.

Limitations and Future Work

This study focused on the three translational degrees of freedom ($X, Y, Z$). Extending the theory to include rotations (roll, pitch, yaw) adds significant complexity regarding moments and contact points. Additionally, the current stability analysis assumes a static environment; future work will need to account for the dynamics of moving or flexible objects. The authors also suggest exploring "energy tanks" to provide even more robust stability guarantees for asymmetric systems.

Summary updated

Abstract

In contact-rich tasks, robots can correct positional errors by generating compliant motion. Methods using compliance control have the advantage of easily adjusting parameters. However, employing an asymmetric matrix as the stiffness matrix extends design freedom while posing a challenge in demonstrating passivity. This study includes a stability analysis in 3-dimensional space based on the root locus method and proposes a parameter design method based on the analysis. The asymmetric part generates a curl force field that may cause instability. Nonetheless, by designing the asymmetric part such that the eigenvalues of the stiffness matrix exclude imaginary components, vibration resulting from the rotational force field can be avoided.
BibTeX
@article{tsuji2024admittance,
  author = {Toshiaki Tsuji and Yasuhiro Kato},
  title = {{Admittance Control of a Multi-degree-of-freedom Robot using an Asymmetric Matrix}},
  journal = {Journal of the Robotics Society of Japan},
  year = {2024},
  doi = {10.7210/jrsj.42.1027},
}
Figure