GPS-denied navigation. Optical flow (downward camera) and visual-inertial odometry (forward or downward camera). Supports ArduPilot, PX4, and iNav 7.0+ on USB UVC or CSI cameras.
Lets the drone hold position when GPS is unreliable — indoors, under bridges, in urban canyons. A camera and the IMU feed the flight controller a 30 Hz position estimate over MAVLink, so position-hold, loiter, and missions keep working unchanged.
Six modes: optical flow with a rangefinder (fastest, low altitude on textured ground), without one (barometer height, drifts above 5 m), OpenVINS VIO (short flights), VINS-Fusion VIO (long loops with revisits), Hybrid (both, ~1.5× CPU), or Off.
Every 250 ms it checks itself. If features drop or drift grows it steps down — VIO → optical flow → barometer hover → land — and warns the GCS at each step.
Auto-detects ArduPilot 4.5+, PX4 1.14+, and iNav 7.0+. Global-shutter USB camera preferred for VIO; rolling-shutter or CSI fine for optical flow. VIO needs a one-time ~10 min camera-IMU calibration. Watch feature count and drift on the Vision Nav panel — drops = lost scene, growth = re-init.
Flight behaviors this extension adds to the cockpit Skill Bar, each bindable to a hotkey or gamepad button.
The capabilities this extension requests. You approve them at install time; the sandbox denies anything not listed here.
Forecast footprint on the companion computer. The supervisor enforces the hard limits.
Live channels this extension publishes to Mission Control.
Open Mission Control, then use the Extensions tab on a compatible node to browse the registry and install this extension. Mission Control installs the signed archive on the node you select; it activates after the signature is confirmed and the requested capabilities are approved.
This extension bundles upstream open-source components. Each component retains its original license and copyright. Source code links below.
VIO mode vio_openvins uses the OpenVINS estimator from the Robot Perception and Navigation Group at the University of Delaware.
https://github.com/rpng/open_vinsVIO mode vio_vins_fusion uses the VINS-Fusion estimator from the HKUST Aerial Robotics Group.
https://github.com/HKUST-Aerial-Robotics/VINS-FusionThe published archive now carries the cross-compiled Rust agent binary, so vision navigation installs as a complete extension. Adds an Engage cockpit Skill that arms and disarms vision navigation from the Skill Bar. No change to the estimator behavior.
Layman-tone four-paragraph description; resource impact card advertises 30 Hz pose output instead of CPU peak; structural drops of MAVLink message names, MSCKF / bundle-adjustment labels, explicit threshold numerics, and calib path so the modal reads in ~30 seconds without losing operator-relevant detail.
Tighter description copy: the ten-paragraph how-it-works narrative is condensed to five tighter paragraphs that preserve the same coverage (what + how, six modes, fallback ladder thresholds, firmware and hardware preferences, calibration and in-flight telemetry) at roughly half the rendered length. No runtime behavior change.
description_long rewritten as a ten-paragraph how-it-works narrative covering the pipeline, optical-flow and VIO mechanics, hybrid cross-check, fallback-ladder thresholds, firmware emission, hardware preferences, calibration wizard contents, and the in-flight telemetry to monitor. No runtime behavior change.
Description copy enriched with mode-by-mode detail, degradation ladder, firmware emission specifics, and calibration wizard walkthrough so the install modal renders a substantive product page rather than a tagline. No runtime behavior change.
Manifest enrichment: adds long-form description, feature list, hardware requirements, resource impact estimates, required FC parameters, telemetry field list, and a documentation URL for the install-modal pre-install summary. No runtime behavior change.
Adds VIO modes backed by signed vendor binaries. Optical-flow modes remain pure-Python.