Autonomy Failed. We Fixed It. Victus builds contested autonomy software that works where GPS doesn’t, on any platform, from the seafloor to low Earth orbit.
getvictus.ai
info@getvictus.ai
PhantomNAV is our patent-pending Machine Learning State Estimator (MLSE) for Contested Autonomy in GPS-denied environments. A sim-to-real machine-learning pipeline trains domain-specific inertial navigation models in high-fidelity simulation, then deploys lightweight inference packages to commodity edge compute. Models broadcast position corrections to the autopilot over MAVLink or other protocol. Single platform across air and maritime today, with subsea, ground, and orbit on the roadmap. Vendor-agnostic. No GNSS reliance at runtime.
Sim-to-Real
VICTUS PhantomNOISE bridges simulation and hardware noise profiles. Models trained in sim transfer cleanly to live platforms.
Multi-Domain
Single platform spans air and marine. Pluggable architecture extends to subsea and ground (roadmap).
Edge-Deployed
ONNX inference on commodity ARM compute. Real-time at 10–50+ Hz with sub-100 ms latency.
Autopilot-Native
MAVLink position-correction broadcast. ArduPilot primary, PX4 supported. Open Architecture & vendor-agnostic.
Configure
Ontology-driven mission and scenario configuration on enterprise data platform.
Simulate
PhantomSIM runs high-fidelity, multi-domain simulation across air and marine.
Train
PhantomNOISE applies real hardware noise profiles to sim data; temporal ML model trains.
Deploy
ONNX inference package pushed to edge compute (Pi 4/5, Jetson Orin, VOXL, etc).
Navigate
Real-time position corrections broadcast to autopilot via MAVLink. No GNSS required.
Single ML pipeline — pluggable simulator architecture extends to additional platforms without re-architecting the data or training stack.
CURRENT
Air
CURRENT
Maritime
ROADMAP
Subsea
ROADMAP
Ground
ROADMAP
Orbit