⌖ FIELDWORK / NAVIGATION LAB
SYNTHETIC ENVIRONMENT

EXPERIMENT 01 — VISUAL POSITIONING

Finding our way back.

An inertial estimate drifts. A familiar landscape brings it home.

01 REFERENCE LANDSCAPESYNTHETIC SECTOR A · 1000 × 700 m
↗ IN FLIGHTN
↑
100 sim. mE 000 · N 000
Ground truth INS only Visually corrected Camera footprint
OBSERVATION NOTES T + 00.0 s

The camera and inertial estimate begin together. Watch their paths separate.

Camera→Match→Confidence→Correction
0.0 m
0.0 m
— m
0.0 m
No visual fix yet
POSITION ERROR sim. m
— INS only— Visually correctedTime (s)
EXPERIMENT TIMELINE00:00 / 02:00
0 s30 s60 s90 s120 s

THE EXPERIMENT

Two ways to know where you are.

Inertial navigation integrates small motion errors. Over time, its position drifts. A camera supplies an independent observation by comparing the ground with a stored map.

  1. Observe. The live camera samples this exact synthetic landscape.
  2. Search. A historical image index retrieves candidates, then local matching refines pose and checks geometric evidence.
  3. Evaluate. Measured image agreement, spatial verification, distinctiveness and competing candidates determine a heuristic confidence.
  4. Correct. Accepted observations gently pull the fused estimate toward the visual fix. The INS-only path keeps drifting.

Switch environments to see ambiguity and rejection. Compare normalized image correlation, local corners, image-derived geographic relationships, and their hybrid. Benchmark lab replays identical recorded observations and reports failures alongside successes.

All distances are simulated metres. This is an illustrative estimator, not a navigation-grade physical model. Environment, mode, and disturbance changes restart a repeatable experiment.

SENSOR INSPECTION

One landscape. Two observations.

Live sensor and reference imagery

REFERENCE · EARLIER CAPTURELIVE EO · CURRENT OBSERVATION

This pair is held at its capture time. Seasonal growth and sun direction change the rendered world; roads, rivers, building footprints and tree trunks remain fixed. The acceptance score is heuristic, not a probability. It combines the enabled measured image evidence and, for Hybrid, the navigation prior. Ground truth is reserved for evaluation.