The project refactors a ROS 1 single-file prototype into a ROS 2 Humble package with red/blue cone perception, non-blocking PID, chassis command adapters, mission states, camera watchdogs, emergency stop, and explicit fault reset.
Outdoor cone courses bring lighting changes, shadows, occlusion, single boundaries, camera interruption, and chassis protocol differences.
Continuing to drive after target or camera loss turns a perception problem into a safety problem. A single loop mixing perception, control, mission, and ROS communication is also difficult to test and tune.
Each layer owns one core responsibility; parameters live in YAML.
Receive images and monitor startup/runtime timeouts.
Segment colors, filter contours, estimate center and confidence.
Compute steering, limit speed, and handle lost targets.
Mapping, red light, navigation, parking, and safety states.
Publish chassis commands and debug topics from one point.
Dual- and single-boundary estimates use different rules and temporal smoothing.
Steering, speed, mission, and safety state are combined at one output.
Standard Twist and legacy PWM modes remain clearly distinct.
Lights, laps, parking, and e-stop events avoid repeated transitions.