Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

How Swarms Differ: Challenges in Collective Behaviour Comparison

André Fialho Jesus, Jonas Kuckling

TL;DR

The paper tackles the problem of measuring similarity between swarm behaviours, a prerequisite for automatic design, by comparing four literature-based feature sets across four similarity measures and applying a self-organising map for explainable classification. It finds that the performance of similarity assessments strongly depends on the chosen feature set and similarity measure, with some pairings aligning better with perceptual distinctions than others, and that absolute similarity values can be misleading. The SOM-based classification reveals modest accuracy and substantial confusion between visually similar and even distinct behaviours, underscoring the challenge of feature selection for robust discrimination. The work highlights the need for robust, context-aware feature-measure choices and offers SOM-based explainability as a step toward identifying regions in feature space where behaviours cannot be reliably distinguished, informing future automatic swarm design and analysis efforts.

Abstract

Collective behaviours often need to be expressed through numerical features, e.g., for classification or imitation learning. This problem is often addressed by proposing an ad-hoc feature set for a particular swarm behaviour context, usually without further consideration of the solution's resilience outside of the conceived context. Yet, the development of automatic methods to design swarm behaviours is dependent on the ability to measure quantitatively the similarity of swarm behaviours. Hence, we investigate the impact of feature sets for collective behaviours. We select swarm feature sets and similarity measures from prior swarm robotics works, which mainly considered a narrow behavioural context and assess their robustness. We demonstrate that the interplay of feature set and similarity measure makes some combinations more suitable to distinguish groups of similar behaviours. We also propose a self-organised map-based approach to identify regions of the feature space where behaviours cannot be easily distinguished.

How Swarms Differ: Challenges in Collective Behaviour Comparison

TL;DR

The paper tackles the problem of measuring similarity between swarm behaviours, a prerequisite for automatic design, by comparing four literature-based feature sets across four similarity measures and applying a self-organising map for explainable classification. It finds that the performance of similarity assessments strongly depends on the chosen feature set and similarity measure, with some pairings aligning better with perceptual distinctions than others, and that absolute similarity values can be misleading. The SOM-based classification reveals modest accuracy and substantial confusion between visually similar and even distinct behaviours, underscoring the challenge of feature selection for robust discrimination. The work highlights the need for robust, context-aware feature-measure choices and offers SOM-based explainability as a step toward identifying regions in feature space where behaviours cannot be reliably distinguished, informing future automatic swarm design and analysis efforts.

Abstract

Collective behaviours often need to be expressed through numerical features, e.g., for classification or imitation learning. This problem is often addressed by proposing an ad-hoc feature set for a particular swarm behaviour context, usually without further consideration of the solution's resilience outside of the conceived context. Yet, the development of automatic methods to design swarm behaviours is dependent on the ability to measure quantitatively the similarity of swarm behaviours. Hence, we investigate the impact of feature sets for collective behaviours. We select swarm feature sets and similarity measures from prior swarm robotics works, which mainly considered a narrow behavioural context and assess their robustness. We demonstrate that the interplay of feature set and similarity measure makes some combinations more suitable to distinguish groups of similar behaviours. We also propose a self-organised map-based approach to identify regions of the feature space where behaviours cannot be easily distinguished.
Paper Structure (13 sections, 3 figures, 2 tables)

This paper contains 13 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Collective behaviours
  4. Feature sets
  5. Similarity Assessment
  6. Explainable Classification
  7. Experimental Setup
  8. Results
  9. Similarity
  10. Explainable classification
  11. Discussion and Conclusion
  12. Acknowledgments.
  13. Disclosure of Interests.

Figures (3)

  • Figure 1: Similarity assessment for all behaviour pairs. Lines represent the mean scores and the standard deviation over $50$ independent simulations.
  • Figure 2: Analysis of the first seed "Gomes2013" based SOM. Left: the inter-node Euclidean distances. Right: the training classification.
  • Figure 3: Analysis of the first seed SOMs on the training samples. The node colour indicates the true label, and the grey contour indicates incorrect labels. The corresponding inter-node distance maps are available in the supplementary video.