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.
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.
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.
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.
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.
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.
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.