🔢 mathematics

Energy-efficient flocking with nonlinear navigational feedback

This paper generalizes Olfati-Saber's flocking model with nonlinear navigational feedback to establish conditions for exponential velocity alignment and bounded center-of-mass tracking without relying on LaSalle's principle, while also identifying control forces that exclude periodic trajectories and investigating methods to reduce propulsion energy consumption.

Original authors: Oleksandr Dykhovychnyi, Alexander Panchenko

Published 2026-05-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Oleksandr Dykhovychnyi, Alexander Panchenko

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

Imagine a school of fish, a flock of birds, or a swarm of tiny robots moving together as one. They don't have a single leader shouting orders; instead, they follow simple rules to stay together, avoid crashing into each other, and move in the same direction. This paper explores how to make these groups move efficiently, especially when the "rules" they follow are a bit more complex and flexible than the standard ones used in the past.

Here is a breakdown of the paper's findings using everyday analogies:

1. The Problem: The "Rigid" Old Rules

For a long time, scientists modeled these groups using a system called the Olfati-Saber model. Think of this like a strict dance instructor who tells every dancer exactly how far to step and how fast to move, no matter what.

The Issue: This "instructor" is very rigid. It uses linear rules, meaning if you are slightly off course, you get a small push back; if you are way off, you get a huge push back. It's like a spring that pulls you back with the same strength whether you are an inch away or a mile away.
The Energy Cost: This rigidity wastes energy. Even if a robot is only slightly off track, the system might still be applying full force to correct it, draining the battery unnecessarily.

2. The New Idea: The "Smart" Flexible Rules

The authors propose a new model with nonlinear navigational feedback. Imagine replacing that strict dance instructor with a smart coach who knows when to relax.

The "Dead Zone" (Relaxing): If a robot is very close to where it should be (within a small "dead zone"), the coach says, "You're fine, stop pushing!" The robot saves energy by coasting.
The "Panic Button" (Aggressive Correction): If the robot drifts too far away, the coach gets very serious and applies a strong, aggressive correction to get it back on track quickly.
The Result: This flexibility allows the group to "breathe." Small wobbles are handled gently, saving energy, while big mistakes are fixed firmly.

3. The Big Challenge: Proving It Works

In the old, rigid model, scientists could easily prove the group would eventually settle down into a perfect formation because the system always lost energy (like a ball rolling down a hill until it stops).

The New Problem: Because the new "smart" rules turn the "brakes" off when things are going well, the system doesn't always lose energy in a predictable way. It's like a car on a rollercoaster that sometimes coasts and sometimes accelerates.
The Solution: The authors developed a new mathematical method to prove that, despite this lack of a simple "energy hill," the group will still settle into a stable pattern (an "attractor"). They proved that even with these flexible rules, the robots won't fly off into space; they will stay close to their leader and move together.

4. Two Types of "Wobbly" Behavior

The paper discovered something interesting about how these groups behave when they are just cruising along at a constant speed:

The "Still" Flock: If the rules are strict enough (specifically, if the "dead zone" for speed is zero), the group eventually stops moving relative to each other. They form a perfect, static shape behind the leader.
The "Wobbler" Flock: If the rules allow for a "dead zone" (where no correction happens for small speed differences), the group might enter a state where they don't stop moving relative to each other. Instead, they might oscillate or "wobble" in a perfect, repeating circle forever. The authors proved that if the rules are nonlinear (the smart coach), these wobblers generally won't happen for groups larger than two. But if the rules are linear (the strict instructor), the group can get stuck in these endless wobbles.

5. The Energy Experiment: What Saves the Most Battery?

The authors ran computer simulations to see which settings saved the most battery power while keeping the group together.

The Finding: The most energy-efficient setup is not the one used in the classic Olfati-Saber model.
The Sweet Spot:
- Turn off the "Position" force for small distances: Don't try to force the robots to be in a perfect geometric shape if they are already close. Let them drift slightly.
- Rely on "Velocity" force: Use the force that matches speeds to keep them together. If everyone is moving at the same speed, they naturally stay in a loose formation without needing constant energy to push them into a specific spot.
- The "Breathing Room": By allowing a slightly larger "dead zone" for position (letting them drift a bit) and using speed-matching to keep them aligned, the group uses significantly less energy.

Summary

The paper shows that by making the control rules for robot swarms nonlinear (smart and flexible) rather than linear (rigid and constant), we can prove they will still stay together. More importantly, by letting the robots "coast" when they are close to the target and only applying strong corrections when necessary, we can drastically reduce the energy they need to operate. It's the difference between a driver who constantly taps the brakes and a driver who knows when to let the car coast.

Technical Summary: Energy-efficient flocking with nonlinear navigational feedback

Problem Statement
The paper addresses the modeling of collective motion in multi-agent systems, specifically focusing on the conditions required for "flocking"—a state where agents move in the same direction with the same speed. While the widely used Olfati–Saber (O-S) model provides a framework for flocking based on linear navigational feedback forces, it relies on specific assumptions that limit its flexibility for energy-efficient control. Specifically, the O-S model assumes linear feedback forces, which prevents the implementation of control strategies that relax or disable propulsion when deviations from a target trajectory are small. Furthermore, the O-S model relies on a decoupling of the center of mass dynamics and structural dynamics, a property that generally fails when nonlinear navigational feedback forces are introduced.

The authors aim to generalize the O-S model by incorporating nonlinear navigational feedback forces. The primary challenges identified are:

The resulting coupled dynamics are generally not dissipative, making standard Lyapunov-based proofs (like LaSalle's principle) inapplicable for proving the existence of an attractor.
The lack of decoupling complicates the analysis of whether agents will flock to the center of mass and how that center of mass relates to the virtual leader's trajectory.
There is a need to determine if specific configurations of these nonlinear forces can reduce propulsion energy consumption compared to linear models.

Methodology
The authors propose a generalization of the O-S model using the Dissipative Particle Dynamics (DPD) framework to simulate interactions mediated by an ambient medium (e.g., water for swimming robots). The system consists of $N$ agents governed by Newton's equations of motion, subject to:

Ambient Forces: Conservative repulsive forces (modeling collisions) and dissipative damping forces (modeling viscosity).
Self-Propulsion Forces: Nonlinear navigational feedback forces consisting of a position alignment term ( $u_P$ ) and a velocity alignment term ( $u_V$ ). These forces are defined by generating functions $h(s)$ and $g(s)$ that include tunable activation thresholds ( $r_0, v_0$ ), allowing the control to be "relaxed" (turned off) for small deviations.

The analysis proceeds in two main theoretical directions:

Dissipative Dynamics: The authors analyze the case where the virtual leader moves with constant velocity (zero acceleration). They utilize a moving frame coordinate system and construct a weak Lyapunov function (total mechanical energy) to apply LaSalle's invariance principle. They investigate the existence of "wobblers"—non-equilibrium periodic solutions within the attractor.
Non-Dissipative Dynamics: For cases where the virtual leader has bounded acceleration (making the system non-dissipative), the authors treat the nonlinear forces as bounded perturbations of linear forces. They prove the existence of a global absorbing ball and a global attractor without relying on LaSalle's principle. They further analyze the convergence of agent velocities to the center of mass velocity.

Finally, the authors conduct computational investigations using numerical simulations (velocity-Verlet integration) to explore different motion regimes and solve a constrained optimization problem to minimize total battery drain (energy consumption) while maintaining flock formation quality.

Key Contributions

Existence of a Global Attractor: The paper proves the existence of a global attractor for navigational feedback forces that are bounded perturbations of linear ones, assuming the virtual leader's acceleration is bounded. This is achieved without relying on LaSalle's principle, addressing the non-dissipative nature of the general coupled system.
Analysis of Flocking Nature: The authors demonstrate that under mild conditions, agents' velocities converge exponentially to the center of mass velocity. While this consensus velocity may differ from the virtual leader's velocity, the deviation is bounded and controllable via tunable parameters.
Attractor Description and Wobblers: In the dissipative case, the paper describes the attractor for target trajectories yielding dissipative dynamics. It proves that for a broad class of nonlinear alignment forces (specifically when the position alignment force is nonlinear), the attractor does not contain periodic trajectories ("wobblers"), a phenomenon that LaSalle's principle alone cannot rule out. Conversely, for linear forces, such wobblers (periodic rigid motions) can exist.
Energy-Efficient Configurations: Through numerical optimization, the study identifies that energy-efficient regimes are characterized by a shorter cut-off radius for repulsive potential forces and a longer activation radius for position alignment forces. This allows the system to rely on velocity alignment forces for small perturbations, saving energy.

Results

Theoretical Bounds: Explicit time-independent bounds are derived for the positions and velocities of agents relative to the virtual leader. These bounds depend on the tunable control parameters ( $K_p, K_v$ ) and the bounds on the virtual leader's acceleration.
Convergence Rates: The analysis shows that if the position alignment force is sufficiently "relaxed" (large $r_0$ ) such that it vanishes after a spatial stabilization time, the agents' relative velocities converge to zero exponentially.
Simulation Findings:
- In dissipative scenarios, the system exhibits proper exact flocking (convergence to the leader's velocity) in all regimes.
- In non-dissipative scenarios, proper exact flocking is achieved only when "tight" velocity alignment is applied; otherwise, the flocking is approximate (bounded velocity mismatch).
- The absence of position alignment forces leads to faster convergence to velocity consensus but may result in unacceptable spatial dispersion.
- High group density leads to faster velocity consensus but more volatile spatial configurations due to the competition between repulsive and attractive forces.
Energy Efficiency: The simulations indicate that the standard O-S model (linear forces with $r_0=0$ ) is likely sub-optimal. Configurations where the position alignment force is inactive for small deviations (large $r_0$ ) and the velocity alignment force is active for small deviations (small $v_0$ ) significantly reduce energy consumption. The study notes that for certain high thresholds, the position alignment force can become negligible, with formation maintained solely by velocity alignment.

Significance and Claims
The authors claim that their work provides a more flexible framework for controlling multi-agent systems than the linear O-S model, particularly for energy-constrained applications like micrometer- to millimeter-scale swimming robots or sub-centimeter aerial vehicles.

The paper modestly notes that while the theoretical results regarding the existence of the attractor and the exclusion of wobblers are rigorous, the findings regarding energy efficiency are based on computational investigations. The authors state that the observed trends in energy efficiency appear robust but are "not supported by proofs" and should be considered preliminary. They emphasize that their model allows for arbitrary nonlinearities on bounded sets, making it more general than previous extensions that required specific forms of nonlinearity to construct Lyapunov functions. The work suggests that by carefully tuning the activation thresholds and magnitudes of nonlinear feedback forces, it is possible to achieve high-quality flocking with reduced on-board energy consumption, a capability not fully realized in the original linear O-S model.