01 · STATE
Raw returns
Time-of-flight points, radar bursts, SAR phase histories — sensor outputs in their native form, before any interpretation.
— Layered Combat Intelligence Center
Close the gapbetween sensorand decision.
LCIC trains and deploys machine-learning models on LiDAR, radar, SAR, and EO/IR returns — turning raw sensor data into actionable intelligence at the tactical edge.
— How it works
From returns to recognition.
Every detection traverses four states. Our models compress that path so decisions happen in milliseconds, not minutes.
02 · STATE
Voxel grid
Spatial indexing converts unstructured returns into a regular 3D lattice — the substrate every downstream model operates on.
03 · STATE
Detection
Domain-tuned 3D object detectors classify each occupied region: vehicle, vessel, UAS, structure. Bounded, scored, ranked.
04 · STATE
Track
Associations across frames yield persistent tracks — kinematics, intent estimation, and the residual that closes the kill chain.
01
Raw returns
Time-of-flight points, radar bursts, SAR phase histories — sensor outputs in their native form, before any interpretation.
02
Voxel grid
Spatial indexing converts unstructured returns into a regular 3D lattice — the substrate every downstream model operates on.
03
Detection
Domain-tuned 3D object detectors classify each occupied region: vehicle, vessel, UAS, structure. Bounded, scored, ranked.
04
Track
Associations across frames yield persistent tracks — kinematics, intent estimation, and the residual that closes the kill chain.
Capabilities
Three pillars.
Each grounded in published methods, tuned for the conditions our domain actually presents.
- 01
Multimodal sensor fusion
LiDAR, FMCW radar, SAR, and EO/IR ingested through a unified spatiotemporal pipeline. We treat each modality as evidence, not as ground truth — fusion is calibrated to the environment, not assumed.
tag: fusion · pipeline · calibration
- 02
Detection & ATR
Domain-tuned 3D object detectors built on PointPillars-, CenterPoint-, and transformer-based families. We fine-tune against the operational context our deployments will see — maritime, littoral, austere.
tag: ATR · 3D detection · fine-tuning
- 03
Tactical-edge deployment
Models compiled for on-platform inference under DDIL conditions: limited bandwidth, intermittent compute, contested EM. Quantized, latency-budgeted, fail-graceful by design.
tag: edge · DDIL · latency-budgeted
Modalities
Every return tells a different story.
Our pipelines speak each sensor in its native dialect, then translate. No modality assumed, none privileged.
Spatial substrate
Detections need a world to live in.
Every track our models emit is referenced against a continuously-updated geospatial layer — building footprints, road networks, terrain meshes — built from open and licensed sources by our sister project, Spatial Data Systems Research.
Learn about SDSR OSM / 2026 9B+ Geometries indexed Building footprints, roads, terrain features.
OSM EDITS / MO 300M+ Edits / month Continuous updates from the OSM contributor base.
PIPELINE 5 Sensor modalities LiDAR · radar · SAR · EO · IR.
EPSG 4326 / 3857 / 3395 Projections Coordinate systems handled natively.
— Mission
Close the gap between sensor and decision.
Sensors emit volume. Operators need signal. We build the layer in between.
Built for
Mission contexts we design against.
These are the operating conditions that shape our model architectures, training data, and deployment targets.
- 01
Tactical edge
Sensor-to-decision under DDIL — denied, degraded, intermittent, limited. Inference happens on-platform; the network is not assumed.
- 02
C-UAS & wide-area surveillance
Multi-modal detection at standoff. Small targets, low signature, against a noisy clutter background — the detection problem, hardest case.
- 03
Maritime & littoral domain awareness
Persistent observation across radar, SAR, and EO/IR. Tracks survive horizon transitions, weather, and adversarial RF environments.
Contact
Get in touch.
Working on a sensor-intelligence problem that doesn't fit a vendor catalog? So are we.