Adaptive AgroTech logoAdaptive AgroTechENGINEERING INTELLIGENT AGRICULTURE
FIELD ROBOTICS / INTERACTIVE STUDIO

AUTONOMOUS SYSTEMS · TECHNOLOGY DEMONSTRATOR

Every metre is a decision.

See how a field robot perceives its surroundings, follows a mission, and responds when conditions change.

Navigation Studio64 × 44 m experimental field
SIMULATED ENVIRONMENT
MISSION 01Precision row navigation
RTK FIXEDGNSS + SLAM
Preparing the field…
Drag to orbit · Scroll to zoom · Right-drag to pan
Ready to navigateWaypoint 1 · Awaiting start
GROUND SPEED

0.00m/s

TRACKING ERROR

0.00m

True pose → active route
OBSTACLE CLEARANCE

—m

Minimum sensed · this run
POSE UNCERTAINTY

0.02m

Illustrative model value

Perception

ROBOT-CENTRED LiDAR
First returnsDetected objectsRobot heading

The scan updates as the robot moves.

Path following

LAST 60 SIMULATED SECONDS
Cross-track error (m)Waiting for mission start

Decision log

0 EVENTS
  1. Mission ready. Start the robot to see decisions.

FROM PERCEPTION TO ACTION

Trust comes from understanding the decision.

01 / LOCALIZEWhere am I?

RTK GNSS anchors the mission. A local pose estimate supports continuity when positioning conditions change.

02 / PERCEIVEWhat is around me?

Range measurements reveal obstacles. A moving animal calls for a different response from a stationary rock.

03 / PLAN & CONTROLCan I proceed?

The planner searches for clearance; the controller follows the route. Uncertain or blocked conditions can require a stop.

ENGINEERING TRANSPARENCY

What this model does.

A browser-based, planar navigation model displayed in a real-time 3D agricultural environment. It demonstrates decisions, trade-offs and failure responses.

Navigation

A grid-based A* planner searches a known crop map and incorporates sensed obstacles. The controller uses a look-ahead target to follow the resulting path. Robot footprint and configurable clearance inflate occupied space.

Perception

Simulated 360° LiDAR rays stop at the first discrete obstacle. Crop canopy occlusion is not modeled. Detection uses a geometric model, not learned classification. Animal and worker labels are known simulation types. The stereo camera is visual only.

Localization

RTK and SLAM are illustrative pose-error and uncertainty models. No satellite signals, point-cloud registration, map optimization, corrections service or real SLAM algorithm is running. Uncertainty changes the permitted motion.

Protective behavior

Moving obstacles trigger yielding. The scripted animals also stop before the robot body; they are cooperative simulated actors. A disclosed ground-truth simulation interlock also prevents unsafe contact, so zero contacts cannot establish sensor sufficiency or real-world safety.

What remains outside this model

Terrain dynamics, slope, wheel slip, braking hardware, weather, sensor failure distributions, detection probabilities and functional safety verification. Parameters are demonstration assumptions, not product specifications.

Technical background: Pure pursuit · CMU Costmap inflation · Nav2 Path following · Nav2