Bounded Adaptive Sensitivity Through Bio-Inspired Digital Hormone Regulation for Emotionally Intelligent UAV Traffic Monitoring

Abstract

Recently introduced affect-driven UAV controllers model behavioral sensitivity (α) as a static personality-dependent parameter, overlooking the cumulative influence of prolonged operational context. The Pull-Push Engine (PPE) regulates behavioral responses through bounded temporal integration; however, its effective sensitivity remains fixed during execution, limiting adaptive evolution under cumulative operational exposure. To overcome this limitation, this paper introduces the Digital Hormone Layer (DHL), a bounded neuroendocrine-inspired regulatory mechanism that dynamically modulates the PPE's effective sensitivity α eff. In DHL, three scalar hormones inspired by cortisol-, dopamine-, and oxytocin-regulatory motifs accumulate operational context on a medium timescale. In the evaluated scenarios, the behavioral effect is primarily stress-driven, while reward and operator-engagement channels remain architecturally defined but contribute less prominently. Modulation is constrained within a personality envelope by coefficient construction (Personality Preservation Budget (PPB) ρ = 0.20). On emergency events, the DHL-augmented controller responds up to 1.91× faster under multi-stressor exposure relative to the activation-selectivity (EI-Low, α = 0.495) control (95% bootstrap confidence interval [1.70×, 2.14×]). This indicates the advantage arises from the bounded adaptive DHL trajectory, not from a low steady-state α-value. This interpretation is consistent with the implemented per-step hormone-to-sensitivity coupling, observed as an inverse correlation between accumulated stress and effective sensitivity. Mission-final and total in-flight battery consumption are comparable across the single-agent controllers; no battery-efficiency advantage is claimed. A single-equation multi-agent extension shows that increasing the coupling coefficient reduces inter-agent sensitivity distance from 0.00502 (uncoupled, γ = 0) to 0.00243 at γ = 0.50 (95% CI [0.00225, 0.00261]; a 51.6% reduction; one-way ANOVA F = 82.5, p < 0.0001) while both agents remain within the personality envelope, as evidence of bounded inter-agent coupling; system-level multi-agent properties remain to be evaluated.