Forward Dynamics of Variable Topology Mechanisms - The Case of Constraint Activation

This paper presents a physically meaningful transition condition for solving the non-smooth forward dynamics of variable topology mechanisms, offering both redundant and minimal coordinate formulations and validating them through simulations of joint locking in planar and industrial manipulators.

The Gist

Imagine you are driving a car. Normally, you can steer left, right, accelerate, and brake. Your car has a certain number of "degrees of freedom"—ways it can move. Now, imagine a magical scenario where, while you are driving, the car suddenly decides to lock its front wheels. Suddenly, you can no longer steer. The car's "rules of movement" have changed instantly.

This is exactly what happens in Variable Topology Mechanisms (VTM). These are machines (like robots) that can change their own structure while they are moving. They might lock a joint, engage a new contact point, or switch from sliding to sticking.

The paper by Andreas Müller tackles a very tricky problem: How do you predict what happens to a robot the exact millisecond it changes its rules?

Here is the breakdown of the paper using simple analogies:

1. The Problem: The "Magic Trick" of Physics

When a robot locks a joint (like an emergency brake), it doesn't just stop moving smoothly. It experiences a sudden "jolt."

• The Old Way: If you just tell a computer "Okay, stop moving now," the math gets messy. The computer might think the robot's momentum (its "oomph" or energy of motion) disappears or changes in a way that violates the laws of physics. It's like if you tried to stop a spinning top by just telling it to freeze; in reality, the energy has to go somewhere.
• The Real World: In reality, when a joint locks, the energy is conserved, but it gets redistributed. The parts that can still move might suddenly speed up or slow down to compensate for the part that stopped.

2. The Solution: The "Traffic Cop" Formula

Müller proposes a new set of mathematical rules (a "compatibility condition") to act as a Traffic Cop at the moment of the switch.

When the robot decides to lock a joint, this Traffic Cop checks two things instantly:

1. The Geometry: "Okay, the wheel is locked, so you can't move sideways anymore."
2. The Momentum: "But you were moving fast! Since you can't move sideways, where does that energy go? It must transfer to the other parts that are still free to move."

The paper provides two different "rulebooks" (formulas) for this Traffic Cop:

• Rulebook A (Redundant Coordinates): This is like looking at the whole car from a drone. You see every single part, even the ones that are locked. It's very detailed but computationally heavy (like trying to count every grain of sand on a beach).
• Rulebook B (Minimal Coordinates/Voronets): This is like looking at the car from the driver's seat, focusing only on the parts that are actually moving. It's more efficient and faster, but you have to be careful not to lose track of the locked parts.

3. The "Emergency Brake" Analogy

The paper uses a great real-world example: An Emergency Stop.
Imagine a robot arm is moving fast. Suddenly, a human steps in front of it, and the robot's safety system slams the brakes on the first joint.

• Without the new math: The simulation might show the robot stopping instantly and perfectly, which is physically impossible. It ignores the "kickback" energy.
• With the new math: The simulation correctly shows that when Joint 1 locks, the energy doesn't vanish. It transfers to Joint 2 and Joint 3. They might jerk or swing wildly for a split second before settling. This is crucial for safety! If you are designing a robot that works near humans, you need to know exactly where the robot will end up after an emergency stop to ensure it doesn't hit the human.

4. The Experiments: The Pendulum and the Robot Arm

The author tested this on two things:

1. A 3-Link Pendulum: Imagine a hanging chain of three sticks. He locked the second stick, then the third. He showed that if you use his new math, the energy of the swinging parts stays consistent. If you use the old "naive" math, the energy vanishes or appears out of nowhere, making the simulation look fake.
2. A 6-Arm Industrial Robot: He simulated a real factory robot locking its joints one by one. The results showed that the final position where the robot comes to a halt is different depending on whether you use the new math or the old math.
   • Why this matters: If you are programming a robot to stop before hitting a wall, using the wrong math might make you think it stops safely, when in reality, the "kickback" energy makes it swing past the wall and crash.

The Big Takeaway

This paper gives engineers a better way to simulate robots that can change their own shape or lock themselves up. It ensures that when a robot "decides" to lock a joint, the simulation respects the laws of physics (specifically conservation of momentum).

In short: It's the difference between a cartoon where a character freezes instantly when they hit a wall, and a real movie where the character bounces, spins, and transfers that energy into their body. For robots interacting with humans, getting the "bounce" right is a matter of safety.

Technical Summary

1. Problem Statement

Variable Topology Mechanisms (VTMs) are mechanical systems capable of switching between different kinematic topologies, thereby altering their degrees of freedom (DOF) and mobility. While kinematotropic mechanisms change topology via singularities, VTMs discussed in this paper change topology through constraint switching (e.g., joint locking, stiction, or contact activation).

The core challenge lies in the forward dynamics simulation of these systems. When a constraint is activated (e.g., a joint brake locks), the system undergoes a non-smooth dynamic event. This results in:

• Discontinuous system trajectories.
• Impulsive constraint forces.
• A discontinuous jump in generalized velocities.

Standard numerical integration methods fail at these switching events because they cannot satisfy both the kinematic compatibility (new constraints must be satisfied immediately) and momentum conservation (physical laws must hold across the impulse) simultaneously. Existing methods often simply set velocities to zero or ignore momentum jumps, leading to physically incorrect trajectories and energy errors.

2. Methodology

The paper proposes a rigorous mathematical framework to handle the transition conditions at the exact moment of constraint activation ($t=0$). The approach is divided into two main formulations:

A. General Transition Condition

The author derives a general condition based on the Lagrange equations of the first kind. By integrating the equations of motion over an infinitesimal time interval $\epsilon \to 0$ around the switching event, the author establishes a momentum balance equation.

• Kinematic Compatibility: The velocity after the switch ($\dot{q}^+$) must satisfy the new constraint Jacobian ($J_+$).
• Momentum Balance: The change in momentum is driven by impulsive constraint forces ($\Lambda$).
• Resulting System: A linear system of equations (Eq. 22) is formed to solve for the velocity jump ($\Delta \dot{q}$) and the impulsive forces ($\Lambda$):
  $$\begin{pmatrix} M(q_0) & J_+^T(q_0) \\ J_+(q_0) & 0 \end{pmatrix} \begin{pmatrix} \Delta \dot{q} \\ \Lambda \end{pmatrix} = \begin{pmatrix} U(q_0) \\ -J_+(q_0)\dot{q}^- \end{pmatrix}$$
  Where $M$ is the mass matrix, $U$ is the impulse of external forces, and $\dot{q}^-$ is the pre-switch velocity.

B. Specialized Formulations for Constraint Activation

The paper specifically addresses the common case where additional constraints are activated on top of existing ones (e.g., locking a joint while others remain free). Two specific computational strategies are presented:

1. Projected Motion Equations (Redundant Coordinates):

   • Uses Gauß's Principle and null-space projectors ($N_{J,M}$).
   • Formulated in terms of the full set of generalized coordinates ($q$).
   • Equation (32): Solves for $\Delta \dot{q}$ and the impulsive force of the new constraints only, utilizing a projector based on the persistent constraints. This avoids solving for the full system size if the number of new constraints is small.

2. Voronets Equations (Minimal Coordinates):

   • Uses Voronets equations formulated in terms of a minimal set of independent coordinates ($s$).
   • This reduces the dimension of the problem from $n$ (total coordinates) to $n-m$ (independent coordinates).
   • Equation (37): Derives a transition condition specifically for the independent velocities $\Delta \dot{s}$. This avoids the need to invert the full mass matrix $M$ or compute weighted pseudoinverses at the switching instant, which is computationally advantageous for large systems.

3. Key Contributions

• Physically Meaningful Transition Conditions: The paper presents a novel, rigorous condition that ensures momentum conservation during topology changes. Unlike simple "velocity reset" methods, this approach calculates the exact velocity jump required to satisfy both the new constraints and the conservation of momentum.
• Dual Formulations: The author provides two distinct computational pathways:
  1. A redundant coordinate approach suitable for general control schemes.
  2. A minimal coordinate approach (Voronets) optimized for computational efficiency and avoiding singularities in the projection matrix.
• Handling Successive Activation: The method is explicitly designed for the successive activation of multiple constraints (e.g., locking joint 1, then joint 2, then joint 3), which is common in emergency stop scenarios.
• Assumption of Regularity: The method assumes "regular" topology changes (no kinematic singularities are crossed during the switch), which allows for a smooth transition in the configuration space.

4. Results and Validation

The methodology was validated through two numerical examples using an explicit Runge-Kutta 4 integration scheme:

1. Planar 3R Pendulum:

   • Scenario: A 3-link pendulum where joint 2 locks at $t=0.8s$ and joint 3 locks at $t=1.3s$.
   • Comparison: Simulations were run with and without the momentum-consistent switching.
   • Outcome: Without the proposed method, the momenta of unlocked joints were discontinuous, and the trajectory diverged significantly. With the proposed method, the momenta of unlocked joints remained continuous, and the total energy behavior was physically consistent (jumping only at the impulse, otherwise conserved).

2. Industrial 6-DOF Manipulator (Stäubli RX130L):

   • Scenario: Sequential locking of the first three joints (simulating an emergency stop).
   • Outcome: The momentum-consistent simulation resulted in a different final stopping position compared to the inconsistent method.
   • Significance: The difference in the final position highlights the critical safety implication. In human-machine interaction, predicting the exact trajectory during an emergency stop is vital to prevent collisions with humans or the environment.

5. Significance and Applications

• Safety in Human-Machine Interaction: The ability to accurately predict the trajectory of a robot during an emergency stop (where joints lock sequentially) is crucial for safety certification and collision avoidance.
• Model-Based Control: The derived conditions provide the necessary feed-forward terms for inverse dynamics control schemes in variable topology systems.
• Broad Applicability: While demonstrated on rigid bodies, the framework is extendable to:
  • Flexible multibody systems (modal coordinates).
  • Non-holonomic constraints.
  • Metamorphic mechanisms (e.g., robotic hands with variable topology).
  • Docking maneuvers in aerospace.

In conclusion, Müller's paper bridges the gap between non-smooth dynamics theory and practical engineering simulation, providing a mathematically sound and computationally feasible method to simulate the complex dynamics of mechanisms that change their own structure during operation.