LightCortex sits between a drone's autonomy stack and its flight controller, transforming probabilistic decisions into a certifiable safety case — on the compute you already fly.
Robust visual-inertial estimation delivers accurate position and world understanding in fully GPS-denied environments.
A calibrated confidence signal on every estimate — the system detects degradation before position error grows dangerous.
A Control-Barrier-Function safety filter guarantees safe hover, return, or landing when confidence drops below a certified threshold.
Probabilistic autonomy — classical planners or modern learning-based policies — is powerful but unsafe on its own. LightCortex wraps it in a formally-verified safety envelope, so intelligence and guarantees coexist.
Formal constraint satisfaction — geofence, obstacle avoidance, velocity and altitude limits — that holds even when perception degrades. Guarantees are mathematical, not best-effort.
Our novel contribution: safety margins that adapt in real time to the estimated reliability of perception. The safer the system is unsure, the more conservatively it behaves.
Optimal, constraint-aware trajectory generation running in real time on the edge, tightly coupled to the safety filter.
Drops in as a companion to PX4 and ArduPilot stacks. Embodiment-agnostic across aerial and ground platforms.
LightCortex is software-first, engineered to run efficiently on existing edge SoCs — no new hardware to manufacture or deploy.
Validated against ModalAI VOXL 2 (Qualcomm QRB5165) and NVIDIA Jetson Orin-class devices.
Designed for the tight power and thermal budgets of real flight platforms.
Formal guarantees produce the documented safety case regulators require for BVLOS approval.
