1 — Continuous inertial layer
Fast locally autonomous sensing with explicit bias and drift tracking.
MARS BIBLE — REFERENCE GUIDE
Inertial, stellar and optical navigation: locating yourself when Mars has no GPS
IMU, stars, cameras, orbital maps, visual odometry and sensor fusion: build robust navigation without relying on one reference.
ESTABLISHED FACTACTIVE ENGINEERINGPROSPECTIVE CHOICE
Why this deserves a full guide
No GPS constellation at Mars does not mean a Mars vehicle must get lost. Navigation is assembled from complementary references. Inertial sensors provide rapid continuity; stars provide absolute attitude; terrain imagery gives local motion; orbital maps provide global updates; radio and relays add range and velocity information.
A major trend is moving these functions from mission control onto the vehicle. In February 2026 NASA/JPL reported Perseverance using Mars Global Localization to match a Navcam panorama with orbital imagery and refine its location without human help. This is not Martian GPS; it demonstrates onboard perception reducing dependence on Earth localization.
1. Inertial navigation is essential but drifts
An IMU measures angular rates and specific force at high update rate. Integration propagates attitude, velocity and position without an external signal, which is essential during landing, occultation or between external updates.
But bias accumulates. A simplified constant 0.01 m/s² acceleration bias for 100 s produces about 1 m/s velocity error and 50 m position error. Inertial navigation is a memory of motion, not eternal absolute truth.
2. Stars provide an absolute attitude reference
A star tracker images the sky, recognizes star patterns against an onboard catalog and solves three-axis attitude. NASA lists star trackers as core attitude-determination sensors.
Clear field of view matters. Sunlight, bright bodies, reflections or plumes can disrupt solutions, so IMU propagation bridges temporary losses.
3. Cameras and visual odometry
Wheel rotation alone is unreliable because rovers slip. Visual odometry tracks terrain features between images and estimates relative motion, improving autonomous mobility.
As a relative technique it drifts, so global references are periodically needed to prevent error from growing without bound.
4. Mars Global Localization
JPL’s 2026 system transforms a rover panorama and matches it to orbital imagery, connecting local terrain features to a global map.
The demonstrated performance is impressive but environment-dependent. A successful example should not be turned into a universal guarantee for every terrain and lighting condition.
5. Optical navigation during cruise
Interplanetary cameras can observe Mars, moons, stars or other bodies and measure lines of sight. Changing angles over time carry trajectory information.
A line of sight alone does not magically provide all six state components; it is combined with dynamics and other measurements.
6. Terrain-relative navigation during landing
During descent, cameras can compare observed terrain with maps to update position relative to hazards and targets under strict time constraints.
The system must also be able to reject a bad match. Honest uncertainty is safer than a confident false localization.
7. Sensor fusion
Estimators combine inertial propagation with external updates. The IMU predicts; star or terrain observations produce residuals; uncertainty controls correction weight.
Kalman filters formalize this predict-correct logic, allowing a fast drifting source and a slower absolute source to complement each other.
8. Time is the invisible sensor
Camera, gyro and radio measurements must refer to a common time. Timestamp errors become state errors when the vehicle is moving.
Distributed systems therefore synchronize clocks and retain acquisition timestamps rather than only packet-arrival time.
9. Local maps and frames
A settlement needs common frames for habitats, mines, roads, depots and exclusion zones. Orbital maps provide global structure while surface surveys improve local detail.
Positioning becomes shared infrastructure so rescue, construction and transport systems can understand one another.
10. Dust, glare and weak texture
Dust reduces contrast, glare can saturate cameras and featureless plains are hard to match. IMUs also change with temperature and ageing.
Loss of vision, star-tracker outages, inertial bias and relay failure must be preplanned degraded modes.
11. From robots to people
Humans, pressurized rovers and cargo robots may share maps and timing but need different interfaces. Uncertainty must be visible to operators.
Navigation automation therefore needs explainability during incidents: teams must understand why the system believes its current location.
12. Toward Mars PNT
Position, Navigation and Timing is broader than coordinates. A settlement can progressively deploy surface beacons, orbital relays, timing references and map services without copying Earth GPS exactly.
Layered services improve resilience: onboard autonomy first, local beacons where valuable, then increasingly regular orbital infrastructure.
18. When sensors disagree
Sensor fusion becomes most important when sources conflict. A wheel slips while wheel odometry reports motion; vision sees almost none. A gyro reports rotation while a star tracker reports stable attitude. Software must decide whether the discrepancy is normal noise, a failed instrument, temporary obscuration or a bad model.
Responses can include reducing a sensor's weight, declaring it invalid, switching estimators or stopping the vehicle. These decisions need traceable logic so human operators can understand why autonomous behavior changed.
17. Surface navigation without a Mars GPS
Mars has no GPS-like global satellite constellation today. Rovers combine wheel odometry, inertial sensing, vision and ground-team support. Perseverance's 2026 Mars Global Localization demonstration added a powerful idea: autonomously match surface imagery with orbital imagery to reset accumulated localization error.
A future settlement would still need an architecture around that capability. Local beacons, orbital navigation payloads, precision maps and standardized reference frames could together form a more robust regional positioning service.
16. Optical navigation means recognizing geometry
A camera can do much more than take pictures. Landmarks, crater edges, horizons, stars or known bodies provide geometric constraints. On Mars, local imagery can be matched to orbital maps; in cruise, celestial observations can constrain direction and position. The challenge is separating stable features from shadows, dust and illumination changes.
Optical navigation therefore needs algorithms, maps, metadata and computation. Lens geometry, camera calibration and accurate exposure timing matter. A visually excellent image can still be a poor geometric measurement if the camera model is wrong.
15. Kalman filtering: the idea before the matrices
A Kalman filter is often introduced as intimidating matrix algebra. The core idea is simpler: predict what should happen using a model, observe what seems to happen using sensors, compare them, then correct the state according to how much confidence is assigned to the model and each measurement. A noisy measurement should move the estimate less than a high-quality one.
Modern variants handle nonlinear dynamics and richer states, but the teaching principle remains: navigation is not a bag of independent sensors. It is a coherent estimate built from imperfect information, and uncertainty is part of the result rather than an embarrassment to hide.
14. Bias, noise, scale factor and misalignment
Calling a sensor accurate is not specific enough. Gyroscopes can have constant bias, random noise, scale-factor errors and slight axis misalignment. Accelerometers have comparable error families. Some are calibrated before flight; others vary with temperature, vibration and aging.
Serious navigation therefore carries uncertainty estimates along with the state. Two vehicles may display the same numerical position while having very different confidence. A decision to cross hazardous terrain should use that confidence, not only the displayed coordinates.
13. From an IMU to a local map frame
An IMU does not directly output a position on a map. It measures rotation and acceleration in its instrument frame. Navigation software must know that frame orientation, remove estimated biases, handle gravity in the appropriate model, integrate acceleration into velocity and integrate velocity into position. Persistent errors grow through each integration step.
This is why inertial navigation is invaluable during periods without external references but should not be mistaken for absolute truth. It provides continuity between updates; other sensors prevent drift from dominating.
23. Minimum operational glossary
IMU means inertial measurement unit. A gyro senses rotation. An accelerometer senses specific force. Odometry estimates motion from vehicle movement or images. An update or reset corrects a drifting estimate using an external reference.
A reference frame defines coordinates. Bias is systematic offset. Covariance describes uncertainty and correlation. A landmark is a recognized visual feature. Sensor fusion combines several sources into one more robust estimate.
22. Navigation in Martian dust
Dust can degrade cameras, reduce contrast and obscure landmarks, while illumination and shadows change visual appearance. Algorithms proven on clean images need testing under degraded conditions.
Design responses include cleaning, camera redundancy, non-optical sensors, multi-season maps and confidence thresholds. Sensor availability must be treated as a mission variable.
21. Collaborative navigation
Vehicles can help one another. A well-localized rover can become a reference, fixed beacons can broadcast surveyed positions, and surface stations can measure time of flight or arrival angle. This reduces dependence on a single localization source.
Collaboration can also propagate common error. A moved or miscalibrated beacon may corrupt many users, so references need monitoring and exclusion logic.
20. Maps, frames and data versions
A map has resolution, date, coordinate frame and uncertainty. If a rover matches camera imagery to orbital maps, software must know which map version is used and how coordinates transform into the rover local frame.
A settlement changes its own terrain with roads, excavation and construction. Autonomous vehicles using obsolete maps can make internally consistent but physically dangerous decisions.
19. Numerical intuition: small integrated error becomes large
Suppose, only for scale, that a velocity estimate keeps an average error of 0.1 metre per second for 10 minutes. Ten minutes is 600 seconds, so position error can grow to roughly 0.1 × 600 = 60 metres if nothing corrects it. This is not a complete inertial model; it illustrates integration.
The example explains why absolute updates matter. An error that seems tiny over one second can become dangerous after hundreds or thousands of seconds.
27. Navigation quality must be visible to planners
Mission planning software should not treat position as a perfect point. It can use uncertainty regions, confidence levels and expected sensor availability when choosing routes. A narrow corridor between hazards may be acceptable with strong localization but unacceptable after a long period of inertial-only propagation.
This connects estimation directly to operations. The safest route is not always the shortest route; it is the route whose geometry remains compatible with the navigation quality expected along the way.
26. Infrastructure can improve navigation over time
Early missions must carry most navigation capability onboard. A growing settlement can add surveyed landmarks, radio beacons, local maps, orbital navigation payloads and shared correction services. Each infrastructure layer reduces uncertainty for many users and can lower the burden on individual vehicles.
The architecture should still preserve fallback capability. A rover that becomes helpless when one city beacon fails has traded one dependency for another. Infrastructure should improve performance without turning ordinary outages into total loss of navigation.
25. Terrain-relative navigation and hazard maps
Localization answers where the vehicle is; safe navigation also needs to know what surrounds it. Digital elevation models, slope estimates, rock maps and keep-out zones transform a position estimate into operational meaning. A position accurate to one metre can still be unsafe if the hazard map is wrong or stale.
Future Mars infrastructure should therefore maintain both localization maps and hazard layers. Construction, excavation and dust can change local conditions, so map maintenance becomes a continuing municipal engineering function rather than a one-time mission product.
24. Calibration is an operational activity, not only a factory step
Sensors are calibrated before flight, but Mars operations can reveal temperature-dependent offsets, mechanical settling and aging. Some calibration parameters can be updated using known maneuvers, surveyed landmarks or periods when another sensor provides a stronger reference. The calibration state must therefore be versioned and traceable.
A dangerous system quietly changes calibration without preserving history. A robust one records which parameters were active for every navigation solution, so an anomaly can be reconstructed and old science data can be reprocessed if a better calibration later becomes available.
EXPERT LAYER — SYSTEM ARCHITECTURE
Never depend on a single way of knowing where you are
Robustness comes from diversity: inertial continuity, stellar attitude, visual terrain cues, global maps, radio geometry and common time to make all observations compatible.
2 — Multiple absolute references
Stars, terrain, orbital maps and radio update different parts of the state.
3 — Visible uncertainty
The system reports quality, not only a numerical position.
4 — Shared maps and time
Settlement vehicles need common conventions to exchange state without ambiguity.
Check before depending on the system
- camera and IMU calibration
- common timestamping
- versioned reference maps
- false-match detection
- vision-loss modes
- relay-loss modes
- operator-visible uncertainty
What this changes for a real Mars city
A Mars city will need a local mapping and timing service comparable in importance to digital road infrastructure. Construction changes terrain, new sites extend maps, and incidents provide data for improving robustness. Navigation becomes a maintained collective asset.
A good architecture does not promise a perfect position. It promises an estimated position, known uncertainty, multiple ways to cross-check it, and a clear procedure when sensors disagree.
Learn the calculations in Space Academy
Institutional and primary sources
Go further with Arcadia
This public guide stands on its own. Arcadia — Manual of the First Martian City develops these systems as an integrated city architecture.