01
Drone OS
Distributed runtime for perception, fusion, consensus, and safety partitions — designed for contested links, not lab Wi-Fi.
<200 ms
End-to-end decision latency
>85% mAP
Detection accuracy
36 mo
Programme duration
20 UAVs
Physical swarm demonstrator
<20 MB
On-device model size
Positioning
Where we lead
02
Drone software
Fleet-level software: mesh decision protocols, mission logic, and operator-facing integration without central SPOF.
03
Algorithms
Tiny AI detection/classification, decision-level fusion, BFT gossip, and self-defence behaviours — coded to proposal KPIs.
04
Solutions
Deployable packages: platforms, mission pilots, and licensable IP for OEMs and EU public-sector programmes.
Architecture
Software meets the three-layer compute stack
Why centralised drone software fails
Ground-centric drone OS designs depend on continuous high-bandwidth uplinks, create a single point of failure under electronic warfare or GPS-denied conditions, and cannot ship coherent fleet-level algorithms when the link drops. Commodity autopilot firmware alone is not a swarm operating system.
Our approach: decision-level swarm OS
SWARMind is the software and algorithm layer each airframe runs: local Tiny AI, multi-chip fusion, and compact decision vectors (≤512 bytes/drone/s) over mesh. Swarm intelligence is an OS capability — Byzantine-tolerant gossip consensus — not a cloud dashboard bolted onto drones.
L1
AI inference
On-device algorithms (YOLOv8n / MobileNetV3) on Jetson Orin NX-class silicon — <100 ms inference, <20 MB INT8 models.
L2
FPGA fusion
Deterministic fusion runtime on Xilinx Zynq UltraScale+ — EKF in fabric at <1 ms, part of the drone OS real-time path.
L3
MCU safety
Safety partition of the OS: STM32H7 watchdog and Return-to-Home if higher software layers fail — graceful degradation by design.
Featured
Platforms & solutions
Drone platforms and licensable software solutions from the SWARMind stack.
drone
SWARMind Hexacopter Node
Custom ~680 mm hexacopter airframe with multi-chip cognitive payload (~850 g) for Mission swarm sorties.
- Payload budget~850 g
- Endurance KPI>20 min
- ComputeOrin NX + Zynq + STM32H7
- GNSSGalileo / EGNOS
solution
Tiny AI Edge Model Suite
YOLOv8n + MobileNetV3-Small with INT8 quantisation and hardware-aware NAS for Mission-specific detection.
- BaselinesYOLOv8n / MobileNetV3
- QuantisationINT8 PTQ
- Size KPI<20 MB
- Accuracy KPI>85% mAP
solution
Decision Fusion & BFT Mesh
Gossip decision vectors with Byzantine-tolerant quorum for contested and lossy links.
- Vector rate≤512 B/s
- BFT (N=20)f ≤ 6
- Packet loss≥40%
- Node failure≥30%
drone
Multi-Chip Avionics Stack
Three-layer heterogeneous compute: AI inference, FPGA sensor fusion, MCU safe-landing autonomy.
- AI latency<100 ms
- Model size<20 MB INT8
- EKF cycle<1 ms
- DegradationAI→FPGA→MCU
Field readiness
From hangar bay to contested airspace
Physical swarm demonstrators, mission pilots, and licensable drone OS / algorithm IP — one coherent software stack.
Evidence
Why operators choose this stack
- Gate-validated 36-month roadmap for a production-grade drone OS path
- Horizon Europe Missions / EDF dual-track — sovereign European swarm software
- Literature-grounded algorithms (ROS 2, PX4 integration, BFT consensus)
- Software KPIs: <200 ms · >85% mAP · <20 MB INT8 · ≥40% packet-loss resilience
Build on the SWARMind stack
For procurers, UAV/defence OEMs, and research partners who need a leader in drone software, drone OS, algorithms, and deployable solutions — with verified specs, not marketing claims.