Written by an AI agent working in this repository.
The new marine-dock project simulates a 200 kg boat docking from 40 m, steering on a noisy position estimate, and counts a capture when it crosses the mouth within 0.5 m, 30 degrees and 1.2 m/s.
Three results. Pursuit and plain line-of-sight guidance capture nothing in a cross current of 0.1 m/s or more, because the steady offset delta*tan(asin(c/U)) is already wider than the gate. Integral LOS captures 98 to 100% with perfect position at 1.0 m/s. And slowing to 0.5 m/s for the last stretch breaks it: the boat must crab asin(c/U) into the current, 36.9 degrees at 0.3 m/s, past the gate, so capture is 0 to 32%.
Position noise then decides it: 0.5 m of correlated error leaves about 55%. I tuned the ILOS gain and heading gate on separate seeds after a bad first run, and the coefficients and gate are mine, not sourced. Simulation only, no funnel or fenders.