A geometric condition for robot-swarm cohesion and cluster-flock transition

Mathias Casiulis, Eden Arbel, Yoav Lahini, Stefano Martiniani, Naomi Oppenheimer, Matan Yah Ben Zion, Zion, Matan Yah Ben

Abstract

We present a geometric design rule for size-controlled clustering of self-propelled particles. We show that active particles that tend to rotate under an external force have an intrinsic, signed parameter with units of curvature which we call curvity, that can be derived from first principles. Experiments with robots and numerical simulations show that properties of individual robots (radius and curvity) control pair cohesion in a binary system, and the stability of flocking and self-limiting clustering in a swarm, with applications in meta-materials and in embodied decentralized control.

Introduction

In swarm robotics, achieving collective behavior—such as moving in tight formation or huddling together—typically necessitates complex algorithms, high-fidelity sensors, and constant inter-robot communication. This "top-down" approach is computationally expensive and scales poorly to thousands of units. This research explores a "bottom-up" alternative: designing the physical properties of a robot so that its natural interactions lead to desired group behaviors. While the behavior of simple particles at equilibrium is well-understood, "active matter" (self-propelled entities like robots or bacteria) operates far from equilibrium and is significantly harder to predict. This paper addresses this gap by providing a first-principles geometric design rule for robot cohesion.

Methods

The study integrates theoretical physics, numerical simulations, and physical robotics experiments:

  1. Theoretical Modeling: Effective equations of motion were derived from a Newtonian description of vibration-driven robots. This identified a new intrinsic, signed parameter: curvity ($\kappa$), which describes how a particle's trajectory curves in response to external forces (e.g., collisions). Curvity is determined by mass distribution and pivot point locations.
  2. Robotic Experiments: "Bristlebots" (vibration-driven robots) were constructed. Their radius ($b$) was varied using 3D-printed skirts, and their curvity ($\kappa$) was adjusted by changing leg orientation.
  3. Numerical Simulations: Langevin dynamics simulations of up to 8,192 particles were used to map a "phase diagram," correlating curvity, radius, and swarm density with collective states.

Key Insights: The Geometric Criterion

The central finding is a simple geometric inequality: $\kappa + 1/b < 0$. If the sum of a robot's curvity and its physical curvature ($1/radius$) is negative, two robots will naturally "attract" each other upon contact. Curvity acts as a "charge": positive curvity causes a robot to align with external forces (leading to flocking), while negative curvity causes anti-alignment (leading to clustering).

Figure 2: (a-b) Phase portraits showing attraction basins for different κb values. (c) Experimental snapshots of robot pairs. (e) "Kissing time" divergence as the geometric condition is met.

When this geometric condition is satisfied, the "kissing time"—the duration two robots remain in contact—increases by over two orders of magnitude. Unlike standard Motility-Induced Phase Separation (MIPS), where clusters grow indefinitely, these curvity-controlled clusters have a stable, predictable radius $R_c \approx -1/\kappa$.

Key Results and Collective Phases

Experiments and simulations validated the $\kappa + 1/b < 0$ rule and identified three distinct steady states:

  • Flocking: Particles move collectively in one direction (occurs at positive $\kappa$).
  • Active Fluid: Particles move randomly with transient clustering.
  • Arrested Clusters: Particles form solid-like, crystalline clusters with sizes controlled by $\kappa$.
Figure 4: (a-b) Phase diagrams showing the transitions between Flocks, Clusters, and Active Matter (AMB) based on density ϕ and κb. (c-d) Time-evolution of cluster formation.

The researchers demonstrated a "computation-free" control strategy by toggling the sign of $\kappa$ (e.g., by rotating robot legs), triggering transitions between disordered fluid, clustered, and flocking states.

Discussion and Future Work

The current model is limited to 2D environments and assumes circular robot geometries. Future work will explore 3D environments (e.g., drone swarms) and the impact of non-circular shapes. While the derivation is linked to vibration-driven hoppers, the authors suggest curvity is a general property of "dry" active matter. This research paves the way for "active meta-materials" that can dynamically alter mechanical or optical properties by toggling the curvity of their constituent particles.

Summary updated

Abstract

We present a geometric design rule for size-controlled clustering of self-propelled particles. We show that active particles that tend to rotate under an external force have an intrinsic, signed parameter with units of curvature which we call curvity, that can be derived from first principles. Experiments with robots and numerical simulations show that properties of individual robots (radius and curvity) control pair cohesion in a binary system, and the stability of flocking and self-limiting clustering in a swarm, with applications in meta-materials and in embodied decentralized control.
BibTeX
@inproceedings{casiulis2025geometric,
  author = {Mathias Casiulis and Eden Arbel and Yoav Lahini and Stefano Martiniani and Naomi Oppenheimer and Matan Yah Ben Zion and Zion, Matan Yah Ben},
  title = {{A geometric condition for robot-swarm cohesion and cluster-flock transition}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2025},
  doi = {10.1073/pnas.2502211122},
}
Figure