Перейти к содержанию
Главная/Аналитика/Aegisky/GPS-Denied Drone Navigation: Vision, SLAM and Companion Computers for Autonomous Operations
FeatureAegisky

GPS-Denied Drone Navigation: Vision, SLAM and Companion Computers for Autonomous Operations

An engineering explainer on how drones navigate when GNSS is unavailable, covering visual-inertial odometry, LiDAR and vision SLAM, and the companion computers that run them, without model-specific claims.

Aegisky Engineering Desk•20 сентября 2026 г.•7 мин чтения
GPS-Denied Drone Navigation: Vision, SLAM and Companion Computers for Autonomous Operations

How a Drone Navigates When GPS Is Not Available

Most drones rely on satellite positioning to hold position and follow a mission plan. That works outdoors in open sky, but it degrades or fails in environments where the signal is weak, reflected or deliberately disrupted: under tree canopy, inside structures, urban canyons, and areas where satellite signals are jammed or spoofed. GPS-denied navigation is the set of techniques that lets an aircraft estimate where it is and move safely without trusting a satellite fix.

The core idea is to fuse other sensors — cameras, inertial measurement units, and sometimes laser rangefinders or LiDAR — into a continuous estimate of position and motion. This explainer walks through the main approaches, the sensors and the onboard compute they require. Builders sourcing hardware can start from machine vision cameras, autopilots and lidars.

Why GPS Fails and What It Costs

Satellite positioning needs an unobstructed view of multiple satellites. It is vulnerable to several failure modes:

  • Multipath and obstruction. Signals bounce off walls and tree trunks, or are blocked entirely, so the reported position drifts.
  • Jamming and interference. Radio noise on the satellite bands degrades or denies the fix.
  • Spoofing. False signals can induce a wrong position, which is a particular concern in contested environments.
  • When the satellite fix is lost, a flight controller that only trusts GNSS cannot hold position. A GPS-denied stack gives the aircraft an independent position estimate so it can continue its mission, hover accurately, or execute a safe recovery even when the satellite link is gone.

    Visual-Inertial Odometry: Cameras Plus IMU

    The most common GPS-denied technique is visual-inertial odometry, or VIO. It tracks features in consecutive camera frames and combines that visual motion estimate with the aircraft's inertial measurement unit, which measures acceleration and angular rate. The result is a continuous estimate of where the camera has moved, even without any external reference.

    VIO works indoors and outdoors, in corridors and under canopy, wherever there is enough visual texture to track. Its limits are motion blur, low light, and feature-poor scenes like uniform walls or open sky, where there are few features to track. A stereo or depth-capable camera, such as the stereo machine vision camera, gives the VIO stack both image and depth information, which improves robustness.

    SLAM: Building a Map While Localizing

    Where VIO estimates motion along a path, simultaneous localization and mapping, or SLAM, goes further: it builds a map of the environment while keeping track of where the aircraft is within that map. This is valuable for repeatable missions, inspection of known structures, and for returning to a previously visited point.

    SLAM comes in two main flavors:

  • Visual SLAM uses cameras and works well in texture-rich environments; it is lightweight because it uses the same sensors as VIO.
  • LiDAR SLAM uses laser scanners to build a dense point-cloud map. It is robust in low light and featureless visual scenes, at the cost of heavier and more expensive sensors.
  • Many systems combine them. LiDAR sensors for drone SLAM are collected in the lidars category, while the cameras that drive visual approaches are in machine vision cameras.

    The Companion Computer: Where the Software Runs

    VIO and SLAM are not software a typical flight controller can run. They need real computer vision processing — per-frame feature tracking, depth estimation and sensor fusion — which is handled by a companion computer carried alongside the flight controller. The companion computer runs the navigation stack and publishes position estimates to the autopilot over a telemetry link.

    Choosing a companion computer is a trade between onboard compute, weight, power draw and thermal budget. It must be powerful enough to run the chosen SLAM or VIO stack at a rate that keeps the position estimate fresh, yet light and cool enough to fly. Smaller depth cameras, such as the compact depth machine vision camera, are designed to pair with this class of onboard compute. When you source compute, confirm it can interface with your flight controller and that you have the power and thermal headroom to run it continuously.

    Sensor fusion underneath all of this is what turns raw measurements into a stable estimate. An extended Kalman filter, or equivalent estimator, combines the visual or LiDAR motion data with the IMU and any remaining GPS information, weighting each source by its expected reliability. When satellite signals are healthy, the estimator trusts them; when they degrade, it leans more heavily on the visual and inertial sources. The quality of this tuning, and the time synchronization between the camera frames and the IMU samples, often matters more than the raw horsepower of the companion computer.

    How the Pieces Talk: Autopilot and MAVLink

    The companion computer does not directly fly the aircraft. It produces a position and velocity estimate and sends it to the autopilot, which closes the actual flight-control loop. The standard bridge between them is a telemetry protocol such as MAVLink, over which the companion computer emits position-vision or odometry messages.

    This separation matters for design: the autopilot stays responsible for stabilization and safety, while the companion computer is a removable, upgradable navigation module. If the vision stack fails, the autopilot can fall back to its last-known state or a safe behavior. Pick an autopilot in the autopilots category that accepts external vision or odometry input, and confirm the message interface your software stack supports before you commit to hardware.

    Sensor and Compute Selection by Environment

    There is no single best stack; the right combination follows the operating environment.

  • Indoor corridors, warehouses and structures. Visual VIO or visual SLAM with a stereo camera usually suffices, and it keeps weight low.
  • Low-light or texture-poor scenes. Add LiDAR SLAM, which does not depend on visible features.
  • Canopy and forest. Expect multipath GNSS and uneven visual texture; a vision-plus-IMU fusion that degrades gracefully is preferable to relying on satellite fixes.
  • Contested or jammed environments. Prioritize a fully inertial and visual solution that needs no external signal at all, and plan safe fallback behaviors.
  • Budget for testing in the actual operating environment, because lab performance rarely transfers cleanly to real canopy, dust or vibration.

    Integration Pitfalls to Plan Around

    A GPS-denied stack fails most often at integration, not at the core algorithm. Common issues include:

  • Calibration. Camera-to-IMU and camera-to-body extrinsics must be accurately calibrated; poor calibration drifts the whole estimate.
  • Latency. Delayed position messages cause the autopilot to chase stale estimates. Keep the telemetry link fast and well-timed.
  • Lighting and motion. Fast turns and low light blur frames and break feature tracking; choose exposure and frame-rate settings for the mission.
  • Weight and power. A powerful companion computer adds current draw and heat, which cuts into endurance.
  • Fallback behavior. Decide in advance what the aircraft does when the vision estimate degrades or is lost.
  • Treat the navigation stack as a system, and test its failure modes as deliberately as its nominal performance. Build test scenarios that intentionally degrade or remove each sensor, and confirm the aircraft behaves predictably rather than drifting or crashing.

    Keep a calibrated reference configuration and a set of benchmark test routes so changes to firmware or sensors can be compared against a known baseline rather than judged from impression.

    Frequently Asked Questions

    What does GPS-denied navigation actually mean for a drone?

    It means the drone estimates its position and motion from onboard sensors — cameras, inertial measurement units and often LiDAR — rather than from satellite signals, so it can hold position and fly missions where GNSS is weak, blocked, jammed or unavailable.

    What is the difference between VIO and SLAM?

    VIO estimates motion by tracking visual features and fusing them with inertial data. SLAM goes further and builds a map of the environment while localizing within it, which supports repeatable missions and return-to-point tasks.

    Do I need a companion computer for GPS-denied flight?

    Yes. The computer vision and sensor fusion that VIO and SLAM require run on a companion computer separate from the flight controller, which publishes position estimates to the autopilot over a telemetry link.

    Can LiDAR replace cameras for indoor drone navigation?

    LiDAR SLAM works well in low light and featureless visual scenes, but it is heavier and costs more. Many systems combine LiDAR with vision and IMU fusion for robustness across environments.

    How does the vision system connect to the flight controller?

    Typically over a telemetry link using a standard protocol such as MAVLink, with the companion computer sending position-vision or odometry messages. The autopilot keeps control authority and can fall back if the vision estimate is lost.

    Проверено

    Aegisky Engineering Desk

    UAS systems, avionics & counter-UAS technology

    The Aegisky Engineering Desk explains how UAS and counter-UAS systems actually work, from autopilots and long-range links to propulsion, sensors and RF detection. Content is built from manufacturer specifications and catalog data, with the engineering trade-offs stated rather than hidden.

    • Autopilots & flight controllers
    • RC & video links, link budgets
    • Propulsion, ESCs & power systems
    • Sensors, thermal & C-UAS detection

    Technical claims are traced to manufacturer datasheets and the cited sources, and reviewed for systems-level accuracy. Specs reflect the cited revision and may change with new firmware or hardware revisions.

    Request a quote for this solution

    Fill in the form below and our B2B sales team will provide a tailored quotation within 24 hours.

    Contact details

    Full name *
    Work email *
    Job title / role
    Phone (with country code)

    Organisation

    Organisation type *
    Organisation / company name
    Country / region *
    Organisation website
    VAT / EORI / tax ID
    Intended end use *
    End user / operating unit *

    Your requirements

    Product / solution of interest
    Industry / application
    Estimated quantity
    Budget — min
    Budget — max
    Currency
    Preferred Incoterms
    Preferred payment terms
    Required delivery date
    Required certifications / standards
    Installation, training & after-sales needed
    Current / incumbent supplier (if any)
    Preferred contact method
    How did you hear about Aegisky?

    We reply to the email you provide. Company email speeds up business and controlled-item verification.

    Learn more about Aegisky

    Explore Aegisky's full product catalog on the Aegisky platform.

    View Aegisky Profile