Meet us at DASC 2026

Join Lynx at DASC 2026 for "Containing AI Workloads in Safety-Critical Avionics Using Separation Kernel" and learn how AI can be deployed alongside safety-critical applications using a containment-oriented architecture.

Wednesday, September 16 | 2PM EST | Room: Majestic 4

Design and Validation Framework Applied to an ML-based Embedded System for Automating Helicopter Emergency Stop Maneuvers

September 16, 1 PM - 1:30 PM

Several approaches have been proposed to address AI/ML challenges lying on the intersection of required levels of safety, highest autonomy, and the complexity and variability of the environmental conditions. To support Safe-by-Design and Validation of AI/ML-based systems, we propose an approach that relies upon a Digital Engineering MBSE methodology (DEM), to deliver two contributions: first, the DEM methodology is applied to design an embedded system integrating a Helicopter Emergency Stop (HES) function, featuring a Neural Network (YOLOX) for detection of a leading Helicopter (formation flying), to automate emergency stop maneuvers. Second, an end-to-end digital engineering framework for autonomy is introduced, including safety analysis, ODD-driven ML development, SiL/HiL integration, and simulation- based testing, applied to emulate flying conditions and validate safety/performance requirements. This work outlines perspectives towards certification of ML-based systems in aeronautics, as foreseen by standardization bodies (EASA, EUROCAE, and SAE) and recommended practices in progress.

Containing AI Workloads in Safety-Critical Avionics Using Separation Kernel

September 16, 2 PM - 2:30 PM

AI is rapidly moving into avionics, creating new opportunities for perception, mission management, health monitoring, and decision support. But deploying AI in safety-critical systems also raises complex questions about determinism, interference, assurance, and certification.

In this presentation, Lynx will explore how a Type-1 hypervisor or separation-kernel architecture can contain GPU-accelerated AI inference within a lower-criticality domain—while protecting high-criticality avionics software in a dedicated real-time environment.

Attendees will learn how to:

Minimize shared resources between AI and safety-critical workloads
Constrain interactions through controlled, well-defined interfaces
Preserve determinism in high-criticality avionics software
Support incremental AI adoption in Integrated Modular Avionics platforms

Ethan Salehi

Technical Account Manager, Lynx

Ethan Salehi is an Technical Account Manager at Lynx and a Ph.D. researcher at Kennesaw State University, where his research focuses on safe and secure real-time operating systems for multicore embedded systems. He has extensive experience supporting defense and aerospace programs with certifiable software solutions aligned with MOSA, FACE, and DO-178C standards. Ethan previously served as a FACE Verification Authority at LDRA and contributed to multicore interference analysis initiatives in collaboration with the U.S. Army. His work bridges theoretical architectures and practical deployment of OS-level standardization, supporting mission-critical environments such as degraded visual operations and autonomous systems.