DNAV-1 "GPS-denied navigation" · D-DEMO-PACK-1 "Demo pack" · EUDIS hackathon, 15-17 Oct 2026

Demo run-sheet

Five stops, about five minutes. Each stop: what to open, what to click, one line to say, and the recorded video to play if the live page fails. Every number here is measured in the simulator, marked (sim).

Pages: Jury deck · Results explorer (deck slide 26) · Flight simulator · Results · Demo · Start here · How we test · Index · 14 Oct day summary

Pick a route

Each route has a recorded replay and a live-fly path (D-N3-DEMO-MENU-1). The recordings were flown with preset best (physics + speed AI v3); a live flight defaults to Pathfinder (physics only), the system we lead with (D-N3-LEAD-1). Generated by docs/pitch/build_night3.py.

Before you walk on

  1. 1

    The problem and the idea

    0:00-1:15
    deck
    docs/pitch/deck.html
    1. Title, then the jamming slide, then how it works.
    2. Stop on the progress timeline: dated milestones.

    When GPS is jammed the autopilot loses its position in a minute or two; Pathfinder runs beside ArduPilot and hands it our own estimate instead.

    Fallback

    The deck is a local file and needs no server. Second copy: how-it-works.html.

  2. 2

    Plan a flight in the Control Room

    1:15-2:15
    live
    localhost:8790/#new
    1. Sample run (Cyprus): route around Orounda, jam zone on leg 2.
    2. Next, pick a Setup preset: leave it on best (today the same settings as the reference, baseline-main). Under Systems keep the default, ArduPilot alone and Pathfinder (physics only): the system the demo leads with (D-N3-LEAD-1). Add physics + speed AI to show the learning part.
    3. Next, Next: stop on 4 Review.
    4. Do not press Run on stage: the job waits for the simulator.

    Any route, any jam zone, any wind: the same flight is flown by every system so the comparison is fair.

    Fallback video · 25 s · newrun-flow.webm
  3. 3

    Watch the jam: recorded flight 1002

    2:15-3:45
    live
    localhost:8790/?present=1
    1. If it is not playing: Play a recorded flight.
    2. 5x, then Jump to jam.
    3. Let it run 60-90 s; the labelled tracks pull apart after the red jam mark.

    Same route, same jam: ArduPilot alone ends up more than a kilometre off; Pathfinder stays within a few metres.

    1368 mArduPilot alone, +5 min (sim)
    7.5 mPathfinder, physics only, +5 min (sim)
    8.4 m+ speed AI, +5 min (sim)
    Fallback video · 83 s · replay-1002.webm
  4. 4

    The evidence, including what lost

    3:45-4:45
    page
    docs/RESULTS.html docs/HOW-WE-TEST.html (how we test)
    1. Point at the 1, 2 and 5 min columns.
    2. Then the rows that lost: say them out loud.

    We report what failed too: on flight 1002 the drift-correcting AI made it worse, and terrain matching lost one flight when it steered a full simulated flight.

    85.5 m+ drift AI on 1002, +5 min, vs 7.5 m physics only (sim)
    200 → 128 mterrain matching offline, 300 s (sim)
    1 lostterrain matching, steering a full simulated flight (sim)
    Fallback

    Results is a local file. Published copy: Pathfinder Results.

  5. 5

    How it plugs in

    4:45-5:00
    deck
    docs/pitch/deck.html
    1. Integration slide, then the ask.

    No new autopilot: one process next to ArduPilot. Next step is real flight data; everything so far is simulator.

    Fallback

    Say it without the slide.