Controller Tuning and Robustness for Autonomous Ground Robots in Eucalyptus Plantations: Simulation and Analysis Code

Simulation, tuning and analysis code for the article "Controller Tuning and Robustness for Autonomous Ground Robots in Eucalyptus Plantations". Running it reproduces every number the article reports, from scratch. What it models. Inter-row navigation of a skid-steered platform through a commercial eucalypt stand: stems displaced from the planting lattice, a first-order yaw actuator, collision resolved against the oriented chassis rectangle, and satellite positioning that drops out under canopy and is bridged by a six-state extended Kalman filter. What it runs. Five experiments: a 2x2 factorial ablation of state estimation and terrain disturbance, across two commercial spacings and twelve independent stand realisations under common random numbers; finite-difference descent, Bayesian optimisation and CMA-ES at a matched budget of 40 objective evaluations, against sliding-mode and active disturbance rejection reference controllers; a disturbance amplification sweep, with a breakdown factor per controller; a direct test of whether the closed-loop sensitivity peak predicts operational breakdown, over 48 individually tuned controllers and 8064 missions; the tuners re-run with that margin enforced as a feasibility filter, and a local sensitivity analysis over the five uncalibrated parameters. Running it. python run_all.py --jobs 10, about 45 minutes on ten cores. The pipeline is deterministic given the declared seeds, so a rerun on any machine reproduces the same numbers. Every constant that influences a result is declared in src/config.py; nothing that affects a number is hidden inside an implementation. The result files from the published run are included, so a reader can compare their own run against the reference without running anything first. Scope. This archive is the method. It carries no figure renderer, no table generator and no typesetting: those reproduce the article rather than the study. Python 3.11 with NumPy, SciPy, scikit-learn and pandas, at the versions pinned in requirements.txt.