Skip to content
spacetime / assure

GPS-denied navigation

Test navigation.
Before the mission.

Your routes. Your maps. Your stack.

See the data Bring a mission

Terrain reconstruction · Ojai / CA-33

How much detail is enough?

USGS data & results ↗
Retained grid
Hillshade of the retained 512 by 512 USGS elevation raster
Reference512 × 512 · 262,144 samples
Terrain downsampled to 128 by 128 and reconstructed to the reference grid using bilinear interpolation
Reconstruction128 × 128 · 16,384 retained samples
Fewer height samples
93.75%
RMS height deviation
3.94m
Absolute p95 height deviation
8.92m
Maximum height deviation
33.67m

Terrain representation test. Compared with the retained raster—not surveyed truth or navigation accuracy. Fewer samples does not mean the same reduction in file size.

Measurements ↗Source identity ↗Source SHA-256 verified before rendering

Magnetic navigation · GPS unavailable

Navigate by the field.

Add interference. Watch corrections stop, then recover.

Synthetic · known 1D corridor

Same invented route. Same odometry.

Odometry onlyMagnetically aidedUpdate rejected / warmup
Along-corridor position error through the synthetic replayOdometry error grows during GPS loss. Magnetic aiding is refused when its measurement window contains interference. This is an invented one-dimensional corridor, not measured navigation accuracy.Position error · m02505000 s800 s1600 sElapsed time
Odometry error
270m
Aided error
45.8m

Magnetic update rejected

The magnetic pattern does not fit the reference.

Invented map and measurements. Error shown against synthetic truth; field performance remains unmeasured.

Method, limits & sources

A bounded scalar-pattern search along a known corridor. Rejected updates leave odometry running. Raw sensor compensation and unconstrained 2D navigation are outside this demo.

The uncertainty scale has a 100 m floor and is heuristic, not a calibrated confidence interval. No Ojai survey or PNI flight-accuracy claim is represented here.

PNI research ↗Source & method record ↗Replay data ↗

Your next mission

Three questions.

  1. Where does navigation fail?
  2. Which data closes the gap?
  3. Is the improvement worth the cost?
Test your mission ↗

After the mission

Log the failure.
Verify the fix.
Update the fleet.

Every verified correction can improve the next run.

Proposed workflow. Customer data stays within the agreed boundary.

Mission demo

Mission-conditioned coverage

Follow the route.
Find the gap.

Illustrative workflow
No navigation results are plotted.

DEPARTUREDESTINATIONREVIEWTHIS SEGMENT
Illustrative planned routeCandidate gap

Retained Ojai terrain provides visual context. Route, gap and proposed remedies are illustrative, not a georeferenced mission assessment.

Customer reference

Find the limiting segment.

Review where coverage, alignment or observable detail could limit the planned route. A flagged segment is a question to investigate.

Bring your operating plan ↗

Explore tracks, sightlines and uncertainty ↗

What we compare

Same held-out recordings. Same vehicle configuration. Public maps, your maps, then targeted additions. Model adaptation is measured separately.

Compare positioning error, false fixes, recovery, time within limits and onboard cost. Resolution, coordinate accuracy and freshness are separate variables.

Evaluation protocol ↗

Navigation results remain unmeasured. Public references are baselines; absolute accuracy requires independent control.

Data and deployment

Prepare customer maps for an agreed area, sensor set and onboard runtime. Compare corrections, new imagery, collection or a route change. Tasking is scoped with providers. Confirm feasibility, rights, price and acceptance criteria before work.

Verified fleet observations can update regional references. Candidate changes are replayed against failures and held-out missions before release, with versioning and rollback.

Packages and services ↗
Models and simulation

Shared geography. Sensor-specific perception. UAV first; ground, surface maritime, underwater and space require separate integration and validation.

Planned Omniverse / Isaac Sim integration supplies simulation and synthetic data. Spacetime’s training pipeline updates the models. Odyssey is being considered for human-facing generated experiences. Benchmark measurements stay tied to source geometry and recorded evidence.

Reference updates, model releases and runtime releases are versioned and tested separately.

Paid evaluation

Bring one mission.

Maps. Route. Vehicle. Error limit.

Request evaluation Scope ↗

General descriptions only. Arrange mission-data transfer separately.

The decision

  • Where your reference falls short.
  • What to fix. What to keep.
  • The evidence. The cost. The next test.