💻 computer science

Stability analysis of admittance control using asymmetric stiffness matrix

This paper investigates the convergence stability of admittance control in robot arms utilizing an asymmetric stiffness matrix, providing derivation methods, integration strategies, and experimental validation to demonstrate the approach's effectiveness in contact-rich tasks.

Original authors: Toshiaki Tsuji, Yasuhiro Kato

Published 2026-03-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Toshiaki Tsuji, Yasuhiro Kato

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Magic Spring" Robot

Imagine you have a robot arm that needs to do delicate work, like screwing a bolt into a wall or sliding a key into a lock. These tasks are tricky because the robot can't be perfectly precise; it might be slightly off-center.

To handle this, engineers give the robot a "soft touch." Instead of being a rigid, unyielding machine, the robot acts like it's attached to a virtual spring. If it bumps into something, the spring lets it slide or wiggle a bit to find the right spot. This is called Admittance Control.

Usually, this virtual spring is "symmetric." Think of a trampoline: if you push down, it pushes back straight up. If you push sideways, it pushes back straight sideways. It's predictable and safe.

The Problem: Sometimes, a simple trampoline isn't enough. Maybe the robot needs to slide diagonally when it hits a wall, or it needs to rotate slightly to fit into a tight corner. To do this, engineers tried making the spring "asymmetric." This is like a trampoline that, when you push down, also pushes you slightly to the left.

The Catch: While this asymmetric spring gives the robot more freedom, it's dangerous. Because it's not a "normal" spring, it can start to spin or wobble uncontrollably, like a car tire that starts to shimmy at high speeds. The big question this paper answers is: How do we use this powerful, weird spring without making the robot go crazy?

The Core Discovery: The "Spin" vs. The "Stop"

The authors discovered that when you use this weird, asymmetric spring, it creates a force that tries to make the robot spin in a spiral (like a corkscrew).

The Symmetric Spring: Pushes you back to the center. You wobble a bit, then stop.
The Asymmetric Spring: Pushes you back, but also tries to spin you. If you don't have enough "brakes" (damping), you will spin forever.

The paper proves that you can use this spinning spring safely, but you have to follow two strict rules:

Rule 1: The "Real Number" Test (No Ghosts)

For the robot to stop spinning, the mathematical "shape" of the spring must be real. If the spring's shape has "imaginary" parts (a math concept, not a ghost), it means the robot will naturally want to spiral.

Analogy: Imagine trying to walk on a floor that is partly solid and partly made of smoke. If the floor is solid (real eigenvalues), you can walk. If it's smoke (imaginary eigenvalues), you'll fall through and spin. The authors found the exact settings where the floor stays solid.

Rule 2: The "Brake" Calculation (Root Locus)

Once you know the floor is solid, you need to know how hard to press the brakes.

Analogy: Think of a swing. If you push a swing, it goes back and forth. To stop it, you need to grab the chains (damping).
With a normal spring, you know exactly how hard to grab to stop it instantly (Critical Damping).
With this weird asymmetric spring, the "grabbing" force needed is different. The authors created a new formula to calculate the minimum braking force needed to stop the robot from spiraling. If you brake harder than this number, the robot stops safely. If you brake softer, it keeps wobbling.

How They Proved It

1. The Simulation (The Video Game Test)
They built a computer model of the robot.

They set the "brakes" too low: The robot hit the wall and started spiraling out of control (like a car losing traction).
They set the "brakes" just above their new calculated limit: The robot hit the wall, wobbled a tiny bit, and then settled perfectly.
Result: Their math formula worked perfectly. It predicted exactly when the robot would be safe.

2. The Real-World Experiment (The Metal Board)
They put a real robot arm in a lab and made it slide against a metal board.

Normal Spring: The robot hit the board and stopped, but it was a bit off-target (8mm error).
Weird Asymmetric Spring: Because they used their new "brake" settings, the robot slid diagonally exactly how they wanted. It hit the board with more force and was more accurate (only 6mm error).
Result: The asymmetric spring actually helped the robot do a better job, as long as they followed the safety rules.

Why This Matters

This paper is like finding a new gear for a car.

Before, engineers only had "Standard Gears" (Symmetric Springs). They were safe but limited.
Now, they have "Sport Gears" (Asymmetric Springs). These allow the car to corner faster and handle tricky roads better.
The Warning Label: The paper provides the Owner's Manual for these Sport Gears. It tells you exactly how much fuel (braking power) you need to keep the car from flipping over.

In short: You can make robots more flexible and capable by using "weird" springs, but you must calculate the brakes carefully. If you do, the robot becomes a master of delicate, contact-heavy tasks. If you don't, it might just spin out of control.

1. Problem Statement

In contact-rich robotic tasks (e.g., peg-in-hole assembly), compliance control is essential for handling varying mechanical constraints. While Variable Impedance Control (VIC) and Admittance Control are standard methods, they typically rely on symmetric stiffness matrices.

Limitation of Symmetry: Symmetric matrices limit the range of achievable motion induction and force interaction.
Potential of Asymmetry: Introducing an asymmetric stiffness matrix can extend the robot's capability and motion range. However, the asymmetric component creates a "curl field" (non-conservative force) that can induce spiral oscillations.
The Core Challenge: Traditional stability analysis relies on passivity. Because the asymmetric part of the stiffness matrix is non-passive, existing passivity-based stability proofs fail. Consequently, there is no established method to guarantee stability when using asymmetric stiffness, limiting its practical application.

2. Methodology

The authors propose a stability analysis framework based on root locus and eigenvalue analysis rather than passivity.

System Model:
- The study focuses on a 2D planar admittance model: $F_{ext} = M\ddot{x} + D\dot{x} + Kx$ .
- The stiffness matrix $K$ is decomposed into a symmetric part ( $K_s$ ) and an asymmetric part ( $K_a$ ).
- The damping matrix $D$ is assumed to be diagonal.
State-Space Representation:
- The system is converted into a state-space equation $\dot{x} = Ax$ , where $A$ is the system matrix.
- Stability is determined by the eigenvalues of matrix $A$ . If all eigenvalues have negative real parts, the system is stable.
Root Locus Analysis:
- The authors analyze how the poles (eigenvalues) of the system move as the damping ratio ( $d_m$ ) varies from 0 to infinity.
- They investigate the conditions under which the eigenvalues remain real versus when they become complex (leading to oscillations).
Derivation of Stability Conditions:
- The paper derives mathematical inequalities to ensure that the system does not exhibit unstable spiral oscillations.
- It establishes a relationship between the asymmetric stiffness component ( $k_a$ ), the symmetric stiffness components ( $k_x, k_y, k_s$ ), and the required damping ( $d$ ).

3. Key Contributions

The paper makes three primary theoretical and practical contributions:

Clarification of Spiral Oscillation Conditions:
- The authors prove that spiral oscillations are excited when the stiffness matrix has complex eigenvalues.
- They demonstrate that the condition for the absence of spiral oscillation is equivalent to the condition that the eigenvalues of the stiffness matrix are real numbers.
Stability Criteria for Asymmetric Matrices:
- Since passivity cannot be guaranteed, the authors derived specific parameter constraints to ensure convergence stability.
- Condition 1 (No Spiral Oscillation): To prevent the existence of imaginary components in the eigenvalues (which cause spiral paths), the asymmetric element $k_a$ must satisfy:
  $|k_a| < \sqrt{k_s^2 + \frac{1}{4}(k_x - k_y)^2}$
- Condition 2 (Damping Requirement): If the above condition is met, a minimum damping coefficient $d$ is required to ensure all eigenvalues have negative real parts:
  $d > \sqrt{\frac{k_a^2}{k_{min}}}$
  (Where $k_{min}$ is the smaller of the diagonal stiffness terms).
Integration into Admittance Control:
- The paper provides a concrete method to calculate the necessary damping parameters to stabilize an admittance controller using an asymmetric stiffness matrix, effectively extending the design space beyond symmetric constraints.

4. Results

The proposed method was validated through both simulations and physical experiments.

Simulation Results:
- Time Response: Simulations showed that when the derived damping threshold was met, the system converged without spiral oscillations. When the damping was below the threshold, the system exhibited growing or persistent oscillations.
- Passivity Check: The study confirmed that while the system is not strictly passive (energy can increase locally during low-velocity phases due to the curl field), it is globally stable if the derived damping conditions are satisfied.
- Parameter Sensitivity: The simulations verified that increasing the symmetric off-diagonal term ( $k_s$ ) reduces the imaginary component of the eigenvalues, thereby reducing the risk of spiral oscillation.
Experimental Results:
- A robot arm was tasked with moving vertically against a metallic board while following the surface in the Y-axis.
- Comparison: The asymmetric (triangular) stiffness matrix achieved a smaller position error (6 mm) compared to the symmetric matrix (8 mm).
- Force Control: The asymmetric setting allowed for better maintenance of contact force ( $F_z$ ) because the displacement was more effectively directed along the intended axis, minimizing unwanted lateral movement.

5. Significance

Breaking the Passivity Barrier: This work challenges the dogma that passivity is the only route to stability in variable impedance control. It proves that asymmetric stiffness can be used safely if specific eigenvalue-based constraints are met.
Enhanced Performance: By allowing asymmetric stiffness, robots can achieve more versatile and precise interactions in complex assembly tasks, overcoming the limitations of symmetric potential fields.
Practical Design Guidelines: The paper provides explicit formulas for engineers to calculate the minimum required damping for a given asymmetric stiffness configuration, making the technology immediately applicable to real-world robotic systems.
Future Outlook: The authors suggest extending this theory to 3D systems and integrating it with Variable Stiffness Control (VSC) and energy tank concepts for broader applications.

In summary, this paper provides the first rigorous stability proof for admittance control using asymmetric stiffness matrices, offering a mathematical framework to safely exploit the enhanced capabilities of non-passive control strategies.