Submitted:
11 April 2026
Posted:
13 April 2026
You are already at the latest version
Abstract
The proliferation of unmanned aerial vehicles (UAVs) in civil, commercial, and defence domains has exposed a critical architectural gap: existing platforms optimise either communication or perception independently, leaving safety coverage incomplete under simultaneous stress in Beyond Visual Line of Sight (BVLOS) operations. This paper proposes the Risk-Aware UAV Safety Architecture (RASA), a three-layer conceptual framework integrating multi-modal sensor fusion, satellite communication (SATCOM), and AI-driven risk modelling aligned with functional safety principles such as ISO 26262. The RASA framework quantifies operational risk as R(t) = α·U_sensor(t) + β·L_c_norm(t) + γ·U_sensor(t)·L_c_norm(t) — a function of normalised sensor uncertainty and normalised communication latency, with an interaction term capturing compound degradation effects — enabling onboard risk estimation without ground-in-the-loop dependency. Building on prior validated work in multi-modal sensor fusion for safety-critical human detection [10] and SATCOM communication architectures for UAV connectivity [15], this paper extends those contributions to the BVLOS domain. Monte Carlo simulations across three representative operational scenarios validate the risk model’s behaviour and demonstrate that the interaction term produces steeper risk escalation under compound failure conditions compared to the linear baseline. This paper addresses the critical gap in BVLOS UAV safety architectures by integrating perception and communication reliability within a single, auditable, risk-aware framework.
Keywords:
BVLOS UAV
; multi-modal sensor fusion
; SATCOM
; risk-aware architecture
; ISO 26262
; functional safety
; AI decision-making
; RASA framework
; estimator instability
; Monte Carlo simulation
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.
