Northline Robotics — ResearchField Notes & Publications · 00:00:00Z

Field Notes & Publications

We publish the problems we solve.

Field robotics is field robotics. The environment changes — snow, grass, altitude — but the problems are the same: localize, perceive, plan, act, and learn in the real world. This series works through those problems in depth, using autonomous grounds care as the worked domain. Every concept here transfers directly to drones and other robots.

01  Cornerstone Series

The autonomy stack, one problem at a time.

LocalizationPerceptionPlanning
01
Localization & State Estimation

Localizing a robot when GPS won't cooperate

Probabilistic state estimation, the Extended Kalman Filter, visual-inertial odometry, and SLAM — how a robot keeps a centimetre fix under canopy, and how the same math flies a GPS-denied drone.

02
Perception · Segmentation

Teaching a robot where the lawn ends

Wire-free boundary detection with semantic segmentation, projecting pixels to a metric ground map, and turning a learned map into a geofence the planner will never cross.

03
Planning · Coverage

Covering every blade: complete coverage path planning

Boustrophedon patterns, cellular and Morse decomposition, spanning-tree coverage, and the turn-cost objective — the same planning that flies a survey drone over a field.

04
Safety · Perception

Seeing the child before the blade

Detection under occlusion, the latency-and-braking budget, diverse-redundant sensing, and functional-safety standards — the requirement that is not allowed to fail.

05
Control & State Estimation

Staying upright on a wet slope

Slip ratio, terramechanics, rollover stability, and folding traction into the state estimator — keeping wheels, and the whole machine, under control on a deformable grade.

06
Data & MLOps

From mowing runs to models: the data flywheel

Data engineering, auto-labeling, active learning, drift detection, and MLOps — how a fleet's operations compound into better models, and a moat.

Field Notes

New problem, worked end to end — in your inbox.

Occasional, technical, no marketing. For engineers, researchers, and partners.