People on the lawn
The perception layer recognises adults and children near the unit and slows the unit before they enter the cutting line.
If a person stays in the cutting line, the unit holds and resumes once the lane is clear.
The GFLS Fusion Stack
Robot Lawn Mower Technology: The GFLS Fusion Stack Under Every UNICUT Unit
GFLS is the localisation platform under SmartMowBot's UNICUT robot lawn mower line. Three sensor layers run in parallel, each layer reads the lawn in a different way, and the three readings are fused into one localisation decision per cycle. H1, H3 and H3 PRO carry the full GFLS stack with an RTK module on board. H5 runs the pure-vision variant of the same platform, with the RTK layer dropped and the VSLAM and AI vision layers carrying the load.
Read the stack top to bottom, then read the convergence on the right.
RTK · Satellite-anchored position reference
On-board RTK module reads a position reference from satellite plus base-station signal when sky view is clean.
VSLAM · Lawn map from camera and odometry
Camera frames plus wheel odometry build and re-use a visual SLAM map of the lawn across sessions.
AI Vision · 300+ object types, edge intent
On-board AI vision classifies 300+ object types and reads edge intent for obstacle avoidance and slow-down behaviour.
Fused Localization Decision
One fused decision per cycle. The platform arbitrates between layers when one layer reads weak.
GFLS = RTK + VSLAM + AI VISION
Three Layers, One Decision
The GFLS fusion localization stack reads the lawn in three layers and arbitrates in one decision
Three vertical columns describe the three layers in depth. Each column lists the same four blocks across the column body so the layers can be compared block-for-block: sensor input, role in the fused decision, where the layer shines on its own, and where the layer needs the other two. Under the three columns, a fused-decision ribbon shows the three layer arrows merging into one output.
RTK · Satellite-anchored reference
Sensor input
On-board RTK module reading the satellite signal stream and a paired base-station reference.
Role in the fused decision
Carries a satellite-anchored position reference when the sky view above the unit is clean.
Where the layer shines
Open lawns, low fence, no overhead canopy, low building eaves; the RTK fix is stable and tight.
Where the layer needs the other two
Under tree canopy, against tall walls, inside courtyards, near eaves; the RTK fix weakens and the VSLAM map plus AI vision layers carry the cycle.
VSLAM · Lawn map from camera and odometry
Sensor input
Camera frames from the on-board camera plus wheel odometry counts from the drive train.
Role in the fused decision
Builds a visual SLAM map of the lawn and re-uses the map across sessions so the unit knows the yard.
Where the layer shines
Repeat sessions on the same lawn, defined visual features, mowing patterns the platform has already learned.
Where the layer needs the other two
Featureless turf, fresh-cut uniform surface, dewy lens; the AI vision layer reads obstacles, the RTK layer anchors position.
AI Vision · 300+ object types, edge intent
Sensor input
The same camera frames the VSLAM layer reads, passed through the on-board perception model.
Role in the fused decision
Classifies 300+ object types around the unit and reads edge intent so obstacles slow the unit or trigger avoidance.
Where the layer shines
Mixed lawns with pets, garden tools, hoses, low furniture, kids' toys; the model recognises the category and adjusts behaviour.
Where the layer needs the other two
Low light, dusk, dewy turf, low-contrast scenes; the VSLAM map and the RTK reference hold position while the perception layer reads with less confidence.
The three layers feed the fused decision per cycle. The platform arbitrates between layers when one layer reads weak, so the layer that reads strongest in the current condition carries the cycle while the other two stay in the loop.
Obstacle Avoidance Breadth
The AI vision robot lawn mower perception layer reads 300+ object types across six lawn categories
Six lawn-side categories cover the obstacle scenarios brand evaluators name first. Each cell shows the category, a short behaviour line describing how the platform reacts, and the matrix footer ribbon carries the 300+ evidence chip plus the edge-intent behaviour line.
Pets and small animals
Pets and small animals are recognised as moving lawn-side bodies and trigger slow-down plus avoidance.
The unit holds a wider buffer around recognised animal categories than around static furniture.
Garden tools and toys
Spades, rakes, kids' toys and dropped tools on the lawn are recognised and routed around.
If a tool is left in a recurring spot, the visual SLAM map records the spot and the unit re-routes on the next session.
Lawn furniture and planters
Tables, chairs, planters and low decoration objects on the lawn are recognised as static obstacles.
The unit holds an edge offset around static furniture and trims the surrounding lawn without contact.
Hoses, cables and edges
Garden hoses, cables and lawn-bed edges are read as continuous low objects across the lane.
The unit slows along the line of the hose or cable and routes around rather than over the obstacle.
Unknown objects on the lawn
When the perception layer reads an object it does not classify, the default behaviour is slow, hold and route around.
Unknown objects do not trigger contact; the default unknown-object posture is the same as a static furniture posture.
300+ object types
Edge intent read on the fly
Unknown objects get a static posture
Three layers in parallel. One fused decision per cycle.
Walk the Boundary, Keep the Map
How the perimeter-walk virtual boundary works on the GFLS platform
A single top-down lawn-map view describes the virtual-boundary algorithm. The user walks the unit around the perimeter once. The platform anchors the walked path on the visual SLAM map plus the GPS reference, then converts the walked path into a solid virtual boundary stored against the lawn. The boundary stays anchored across sessions, and the app allows keep-out zones, narrow corridors and multi-zone scheduling to be edited without re-walking the perimeter.
01 Perimeter Walk
The user drives the unit around the perimeter of the lawn once. The platform records the walked path as a sequence of camera frames, odometry and GPS samples.
02 SLAM Map Anchor
The platform anchors the walked path on the visual SLAM map of the lawn and reconciles the path against the on-board GPS reference, so the boundary is held by both layers.
03 Boundary Recorded
The walked path is stored as a virtual boundary against the lawn. The boundary stays anchored across mowing sessions. Re-walking is only required when the lawn shape changes.
Keep-out zones · Narrow corridors · Multi-zone scheduling
Keep-out zones, narrow corridors and multi-zone scheduling are edited in the app on top of the recorded boundary. Re-walking the perimeter is only required when the lawn shape changes; everyday edits stay in the app.
Operating Window
All-weather robot lawn mower posture across the standard UNICUT platform
Six operational cells cover the conditions brand evaluators ask about first. Each cell carries one factual claim and one factual back-up sentence. Rain detection, IPX6 housing, acoustic posture, dual-radio connectivity, OTA channel and parent-factory continuity stay aligned across every standard UNICUT SKU.
Rain detect + IPX6 housing
The unit detects rain on the lawn and routes back to the dock under the platform default.
Housing carries the IPX6 rating across every standard UNICUT SKU, so the unit holds up to rain on the way back.
Dusk and dewy turf
The fusion stack arbitrates between layers when contrast drops at dusk or on dewy turf.
The VSLAM map and RTK reference hold position while the perception layer reads with less confidence in low-contrast scenes.
Acoustic profile ≤ 59 dB
Acoustic profile stays at 59 dB or lower across the standard UNICUT lineup.
A quiet acoustic profile keeps evening sessions inside the residential noise window the lawn sits in.
WiFi + Bluetooth radio
Every standard UNICUT unit carries a dual radio with WiFi and Bluetooth on board.
The dual radio carries app traffic for virtual fence edits, keep-out zones, narrow corridors and multi-zone scheduling on the brand-side app.
OTA channel on platform
Every standard UNICUT unit receives platform releases through the OTA channel.
Perception, motion control and firmware updates ride the same OTA channel across every SKU so the platform release stays aligned.
Parent-factory continuity
The platform sits on a parent-factory continuity that brand evaluators can audit.
Sunteam Group, founded 1997, publicly listed 2015, runs the Wuhan facility behind the standard UNICUT line. See the Sunteam Group facility.
For commercial-acreage operation under the same platform release, see the commercial track
Which SKU Runs Which Stack
SKU-to-localisation map across the UNICUT lineup
Four rows map every standard UNICUT SKU to its localisation badge and sensor set. H1, H3 and H3 PRO run the full GFLS fusion stack with an on-board RTK module. H5 runs the pure-vision variant of the same platform, with the RTK layer dropped and the VSLAM and AI vision layers carrying the load.
For specs per model and the standard packaging set, see the full UNICUT lineup in the standard catalog
GFLS · Full Fusion Stack
RTK module · monocular camera · 300+ object perception
Open lawns up to 1500 m² where the sky view is clean and the full fusion stack reads tight against the RTK reference.
GFLS · Full Fusion Stack
RTK module · binocular camera · 300+ object perception
Mid-size lawns up to 800 m² where the binocular camera helps the VSLAM map on featureless turf and the RTK reference still anchors the cycle.
GFLS · Full Fusion Stack
RTK module · binocular camera · 300+ object perception · 4G module
Mid-size lawns up to 1200 m² where 1 cm edge cutting and a 4G connection complement the full fusion stack.
Pure Vision · No RTK
Binocular camera · 300+ object perception · install-free
Smaller lawns up to 500 m² where the buyer prefers an install-free unit and the pure-vision variant reads the lawn without an RTK module.
H1, H3 and H3 PRO carry the full GFLS fusion stack. H5 runs the pure-vision variant of the same platform. No standard UNICUT SKU mixes the two badges on one unit.
License the same GFLS stack on the OEM / ODM trackOpen a Technical Evaluation
Start a SmartMowBot technical evaluation
The evaluation form opens a conversation with the platform engineering side. The intake covers brand context, the lawn types the brand serves, the localisation stack the brand is comparing, and the selection-memo checklist the evaluator wants the SmartMowBot side to answer. The form does not handle quoting; quoting follows after the technical evaluation closes.