Change Intelligence
Detect meaningful change across time while separating true activity from sensor variation, seasonal effects, registration error, and benign background motion.
Deimos-One Starlab // Geospatial Intelligence
A living model of the world in motion.
Starlab develops geospatial intelligence systems that fuse heterogeneous sensing, reason across time, expose deception, and convert changing terrain into operationally useful hypotheses, forecasts, and decisions.
Operating picture Fused
Temporal state Updating
Change field Active
Provenance Preserved
Research mandate
Geography is not a backdrop. It is a changing system.
Conventional maps describe where things were. Operational intelligence must determine what changed, why it changed, what the change implies, and which observation should alter the decision.
Starlab treats the physical world as a temporal, contested, partially observed state space. Imagery, signals, terrain, weather, infrastructure, movement, and human reporting become evidence in a common geospatial model rather than isolated layers.
Build systems that understand place, time, motion, uncertainty, and adversarial manipulation as one connected problem.
Observed state Fragmented, delayed, multi-resolution, and sensor-dependent.
Research problem Infer the changing world between observations.
System output Scene graphs, tracks, change fields, forecasts, and decision cues.
Success condition Reduce surprise without hiding uncertainty or source limits.
Active research domains
From pixels and signals to spatial understanding.
Our geospatial research focuses on the reasoning layers beyond collection: multi-temporal change, pattern-of-life, terrain intelligence, multimodal fusion, denied-environment navigation, and environmental forecasting.
Pattern-of-Life
Model recurring movement, dwell, interaction, and spatial behavior to identify deviations, emerging routines, and hidden relationships across time.
Terrain Reasoning
Convert elevation, surface, land cover, weather, access, and visibility into route, exposure, mobility, and mission-relevant terrain assessments.
Multimodal Fusion
Align imagery, radar, spectral, RF, cooperative signals, weather, terrain, and reporting into one temporal geospatial evidence graph.
Contested Navigation
Develop terrain-relative, vision-based, signal-aware, and uncertainty-bounded navigation methods for degraded, denied, or manipulated positioning environments.
Environmental Foresight
Fuse atmospheric, hydrologic, thermal, vegetation, and surface observations to anticipate conditions that alter movement, sensing, infrastructure, and mission risk.
Sensing modalities
No single sensor sees the whole problem.
Starlab research is sensor-agnostic. Each modality contributes a different slice of state, uncertainty, timing, and failure behavior. The intelligence emerges from how those slices are aligned and challenged.
EO // VIS
Electro-Optical
High-resolution visual structure for identification, mapping, object context, and multi-temporal comparison.
Research // visual reasoningIR // THERMAL
Infrared
Thermal signatures for night observation, equipment state, environmental anomalies, and hidden activity cues.
Research // latent heat stateSAR // RADAR
Synthetic Aperture Radar
Day-night, weather-tolerant surface observation for structure, motion, coherent change, and terrain state.
Research // all-weather inferenceHSI // SPECTRAL
Hyperspectral
Dense spectral signatures for material discrimination, subtle anomaly detection, and environmental state estimation.
Research // material reasoningLDR // RANGE
LiDAR
Precise three-dimensional geometry for terrain, canopy, structure, visibility, and surface-change analysis.
Research // 3D world stateRF // EMITTER
RF Sensing
Spatial and temporal characterization of emitters, interference, communications behavior, and spectrum conditions.
Research // non-visual activityAIS // ADS-B
Cooperative Signals
Broadcast tracks fused with imagery and context to identify inconsistencies, missing behavior, and deceptive reporting.
Research // track corroborationMET // ENV
Environmental Networks
Atmospheric, weather, hydrologic, and terrestrial observations that explain or forecast changes in the operating environment.
Research // condition contextTemporal world models
The operating picture is a hypothesis, not a photograph.
A useful geospatial model preserves what was observed, what was inferred, when each element was last valid, and which alternate explanation remains possible. The scene is continuously revised as new evidence arrives.
Identity test Is this the same object, a replacement, or a false association?
Temporal test Which state is current, stale, predicted, or reconstructed?
Causal test What process best explains the observed spatial change?
Decision test Which uncertainty materially changes the recommended action?
Intelligence architecture
Geospatial fusion begins where ordinary mapping ends.
The hard problem is not rendering layers. It is reconciling sensors with different resolutions, clocks, biases, failure modes, and semantic meanings while preserving enough structure to challenge the result.
01 Every observation retains source, timestamp, geometry, and confidence.
02 Tracks and entities can split, merge, decay, or remain unresolved.
03 Change detection is conditioned on sensor and environmental context.
04 Forecasts specify indicators that would confirm or invalidate them.
Adversarial geospatial intelligence
The map can be wrong on purpose.
Contested environments include camouflage, decoys, spoofed tracks, denied signals, manipulated metadata, sensor saturation, and behavior designed to exploit analytic assumptions.
Collection architecture
Platform-agnostic. Mission-directed. Uncertainty-aware.
Geospatial intelligence should not inherit the blind spots of one platform. Starlab research coordinates orbital, near-space, airborne, unmanned, terrestrial, and partner data around the unanswered question rather than the available sensor.
01 Task collection against the highest-value uncertainty.
02 Cross-cue sensors when one modality produces an ambiguous signal.
03 Use persistence to distinguish activity from isolated observation.
04 Replan when weather, access, adversary behavior, or confidence changes.
Research program model
From geospatial question to validated intelligence system.
Starlab programs begin with a decision the current operating picture cannot support. Research is structured around the observations, spatial reasoning, uncertainty, adversary, and transition path required to close that gap.
The output is a working geospatial artifact, validated model, collection strategy, technical finding, or defensible program decision—not a generic map product.
Define the spatial decision.
Identify the area, actors, time horizon, uncertainty, and action the intelligence must support.
Design the evidence layer.
Select modalities, sources, revisit, calibration, and provenance needed to observe the problem.
Construct the world state.
Build scene graphs, tracks, baselines, terrain models, and competing explanations.
Attack the picture.
Introduce registration error, missing data, spoofing, decoys, changing conditions, and false associations.
Prove what survives.
Test against historical, synthetic, adversarial, and live-like scenarios with explicit acceptance criteria.
Move intelligence into use.
Deliver software, models, interfaces, collection logic, geospatial artifacts, and decision rules.
Program inquiries // controlled intake
Build the world model the mission actually needs.
Engage Starlab on geospatial AI, multimodal fusion, temporal world models, change intelligence, terrain reasoning, contested navigation, collection orchestration, and environmental foresight.