COVERAGE CONTROL · SPACE ROBOTICS & AI

First-Order Boustrophedon Coverage Controller

A lawnmower coverage controller rebuilt from coupled PD into a TURN and DRIVE state machine. Proof that control structure, not gain tuning, was the fix.

TYPE

Simulation · ROS 2 Jazzy + Turtlesim

RESULT

Avg cross-track error 0.760 to 0.027

THE FIX

Structure, not gain tuning

The final trajectory: eleven straight, evenly spaced rows over all 22 waypoints.

Cross-track error held under 0.1 units for the entire run, against a 0.5 requirement.

Velocity profiles showing the TURN and DRIVE decoupling: zero linear velocity while turning, zero angular drift on straights.

Overview

The provided coupled PD controller drew spirals: 0.76 average cross-track error no matter the gains. The diagnosis was structural, so the fix was structural: a TURN and DRIVE state machine that rotates in place before every segment and never steers while turning.

Design & Build

Controller: A finite state machine alternating pure rotation and pure translation, with signed cross-track error computed by vector projection with numerical guards.
Waypoints: The generator was rewritten to produce orthogonal rows with vertical step downs at the edges instead of diagonal transitions.
Instrumentation: A custom ROS 2 performance metrics message, the assignment's extra credit item.

Challenges

Recognizing that no amount of PD tuning fixes a controller that couples heading and translation through every segment change.

Result

Average cross-track error cut 96.5 percent, from 0.760 to 0.027 units, maximum from about 2.0 to 0.10, across all 22 waypoints.

Skills & Tools

ROS 2 JazzyControl DesignState MachinesPythonTurtlesim

CODE & DOCS ↗