APOLLO ROBOTICS · MF-2026-11 · SLAM EXPLORATION

Perceive. Localise.
Navigate with intent.

Operate an explainable browser twin of the team’s ROS 2, LiDAR safety, Nav2 and YOLO exploration. Three environments now carry independent geometry, perception targets and mission routes. A* and D* Lite plan the route; AMCL estimates where the robot is.

Environment Route planner

AMCL

Probabilistic localisationParticle cloud initialised

42% confidence

DIGITAL TWIN

Warehouse escort mission

READY

Manual WASD or arrowsGoal —

Safety envelope clearThe robot may move in either direction.

ROBOT VISION

Dual-camera inference

24 FPS

RGB + detectorheading 0°

Scenario-rendered camera evidence with deterministic browser detections. It demonstrates the perception pipeline and interface; it is not a live physical camera or a deployed YOLO model endpoint.

ROBOT STATE

Telemetry and localisation

00:00

Velocity0.00m/s

Directional clearance—m along motion ray

True pose0°, 0heading / cell

AMCL estimate0:0estimated grid cell

Battery100%

Route progress0%

Directional LiDAR safety gateCLEAR

SLAM + AMCL

Occupancy and route map

FreeOccupiedUnknownExecutedPlannedAMCL particles

ROS 2 ARCHITECTURE

Active autonomy graph

packet 0

/scan
LiDAR

/detections
YOLO semantics

/amcl_pose
localisation

A* / D*
route planner

/safety_gate
directional stop

/cmd_vel
actuation

Sensor observations update the occupancy map and AMCL pose estimate. The selected planner proposes motion, but the directional LiDAR gate remains the final authority before velocity reaches the drive controller.

A* finds a complete route

A* combines distance already travelled with a heuristic estimate to the goal. It is efficient when the map is known and stable, so ARGUS uses it as the baseline global planner.

D* repairs a changed route

D* Lite watches the occupancy map. When a temporary obstacle appears, this browser implementation reuses the mission state and replans from the robot’s current cell instead of restarting the mission.

AMCL estimates the robot pose

AMCL is localisation, not a route planner. A particle cloud represents possible poses; simulated scan agreement concentrates the particles and raises localisation confidence while the robot moves.

The safety fan follows motion

ARGUS now evaluates its rectangular footprint and a directional LiDAR fan, not a large circular exclusion zone. A narrow aisle remains traversable when the body fits; red appears only when the commanded direction is obstructed.

ENGINEERING BOUNDARY

Working browser evidence, stated honestly