[Paper] Emotional Modulation in Swarm Decision Dynamics

Source: arXiv - 2603.09963v1

Overview

David Freire‑Obregón’s paper extends the classic “bee equation” for swarm decision‑making by injecting affective states—valence (positive vs. negative) and arousal (low vs. high)—into an agent‑based model. By letting emotions modulate recruitment and inhibition rates, the work shows how emotional contagion can tip the balance of collective choices, offering a bridge between biological swarm theory and human‑centric affective computing.

Key Contributions

• Emotional extension of the bee equation – introduces valence‑arousal variables that directly scale recruitment and cross‑inhibition parameters.
• Facial‑expression mapping – each simulated agent displays a facial cue derived from its affective state, enabling visual analysis of emotional contagion.
• Three experimental scenarios exploring (1) joint valence‑arousal effects on consensus speed and outcome, (2) arousal‑driven tie‑breaking when valence is equal, and (3) a “snowball” tipping‑point effect once a support threshold is crossed.
• Demonstration of non‑linear amplification – even symmetric emotional setups can produce decisive wins due to intrinsic swarm dynamics.
• Open‑source‑ready framework – the model is released as a lightweight Python/NetLogo implementation, ready for reuse in affective AI and multi‑agent simulations.

Methodology

1. Base model – starts from the bee equation, which describes how agents (e.g., bees) recruit peers to a preferred option and inhibit competing options.
2. Emotional state representation – each agent carries a 2‑D affect vector (valence ∈ [‑1, 1], arousal ∈ [0, 1]).
3. Rate modulation – recruitment rate = r₀ × (1 + α·valence + β·arousal); inhibition rate = i₀ × (1 – γ·valence + δ·arousal). Coefficients (α,β,γ,δ) are tuned to reflect how positive mood speeds up recruitment while high arousal can boost both recruitment and inhibition.
4. Agent interaction loop – at each tick, agents randomly encounter peers, exchange their current option, and update facial expressions based on the new affective state.
5. Simulation settings – runs of 500 agents, multiple initial distributions of valence/arousal, and repeated trials to capture stochastic variability.
6. Metrics – consensus ratio (percentage of agents on a single option), convergence time (ticks to reach 95 % agreement), and emotional contagion index (average pairwise valence correlation over time).

Results & Findings

Overall, the study confirms that emotional asymmetries can bias collective decisions, but the underlying non‑linear swarm mechanics can also override emotions, producing “winner‑takes‑all” outcomes from near‑perfect symmetry.

Practical Implications

• Affective AI for collaborative platforms – designers can embed valence‑arousal modifiers into recommendation engines or crowd‑sourcing tools to steer group outcomes without explicit persuasion.
• Human‑robot swarms – robots equipped with affective displays (e.g., LEDs indicating “mood”) could influence human teammates’ willingness to follow certain plans, useful in search‑and‑rescue or manufacturing cells.
• Social media moderation – understanding how emotional contagion accelerates consensus can inform algorithms that detect and dampen rapid polarization spikes.
• Game design & gamified decision‑making – dynamic facial cues tied to player emotions could be leveraged to create emergent team strategies or to balance competitive matchmaking.
• Organizational decision support – visual dashboards that map team sentiment (valence/arousal) to projected convergence times could help managers intervene before deadlock or premature lock‑in.

Limitations & Future Work

• Simplified affect model – the 2‑D valence‑arousal space abstracts away richer emotional nuances (e.g., dominance, appraisal).
• Homogeneous interaction topology – agents interact uniformly; real‑world networks (social graphs, spatial constraints) may alter contagion pathways.
• Parameter calibration – the mapping from affect to recruitment/inhibition rates is heuristic; empirical validation with human or animal groups is needed.
• Scalability – while the current implementation handles hundreds of agents, extending to thousands or to real‑time robotic swarms will require performance optimizations.

Future research directions include integrating physiological sensors (e.g., heart‑rate variability) to infer arousal in real teams, testing the model on heterogeneous network structures, and applying the framework to mixed human‑AI decision pipelines.

Authors

• David Freire-Obregón

Paper Information

• arXiv ID: 2603.09963v1
• Categories: cs.MA, cs.AI
• Published: March 10, 2026
• PDF: Download PDF