Skip to content
Open access

Budget-Efficient PID Tuning of a Line-Following Differential-Drive AGV Using Extra Trees Surrogate-Assisted CMA-ES

Sep 2026 · International Islamic University Malaysia Engineering Journal · 0 citations · 30 references

Abstract

Physical PID tuning for line-following automated guided vehicles (AGVs) is sample-inefficient because the gain-to-performance relationship is nonlinear and noisy, while every candidate must be executed on the real platform. Conventional population optimizers can therefore spend a limited experimental budget on unproductive trials, and evidence for surrogate-assisted CMA-ES under explicit true-evaluation accounting on embedded AGVs remains limited. This study evaluates an Extra Trees surrogate-assisted CMA-ES supervisory tuner on a single line-following differential-drive AGV. CMA-ES generates a virtual gain population in normalized space, and Extra Trees ranks it using the predicted objective and inter-tree dispersion; only a small, ranked subset undergoes physical evaluation. The objective combines integral absolute error, root mean square error, overshoot, actuation smoothness, and a lost-line penalty. We compared the method with pure CMA-ES, particle swarm optimization, and a genetic algorithm on Circle and 8Shape trajectories, using an identical budget of 57 true evaluations for every method, seed, and trajectory. The proposed method achieved the lowest mean final-best objective on both trajectories and improved on the strongest baseline by 30.8% on Circle and 20.8% on 8Shape. Cross-run distributions, empirical cumulative distribution functions, and time-domain responses support the same ranking. These results show that surrogate screening can improve sample efficiency for PID tuning on the tested embedded platform. ABSTRAK: Penalaan PID secara fizikal bagi kenderaan berpandu automatik (AGV) pengikut garisan tidak cekap dari segi sampel kerana hubungan antara gandaan dan prestasi adalah tak linear serta bising, sedangkan setiap calon perlu diuji pada platform sebenar. Pengoptimum berasaskan populasi konvensional boleh menggunakan bajet eksperimen yang terhad pada percubaan yang kurang produktif, dan bukti bagi CMA-ES berbantukan surrogat dengan perakaunan penilaian sebenar yang jelas pada AGV terbenam masih terhad. Kajian ini menilai penala penyeliaan CMA-ES berbantukan surrogat Extra Trees pada sebuah prototaip AGV pacuan kebezaan pengikut garisan. CMA-ES menjana populasi gandaan maya dalam ruang ternormal dan Extra Trees menyusunnya menggunakan objektif ramalan dan sebaran antara pokok, manakala hanya subset kecil dengan skor terbaik menjalani penilaian fizikal. Fungsi objektif menggabungkan ralat mutlak bersepadu, ralat punca min kuasa dua, limpahan, kelancaran aktuasi, dan penalti kehilangan garisan. Kaedah ini dibandingkan dengan CMA-ES tulen, pengoptimuman kawanan zarah, dan algoritma genetik pada trajektori Circle dan 8Shape menggunakan bajet yang sama, iaitu 57 penilaian sebenar bagi setiap kaedah, benih rawak, dan trajektori. Kaedah yang dicadangkan mencapai purata objektif terbaik akhir terendah pada kedua-dua trajektori serta mengatasi garis dasar terbaik sebanyak 30.8% pada Circle dan 20.8% pada 8Shape. Taburan antara larian, fungsi taburan kumulatif empirik, dan respons domain masa menyokong kedudukan yang sama. Dapatan ini menunjukkan bahawa penapisan surrogat dapat meningkatkan kecekapan sampel bagi penalaan PID pada platform terbenam yang diuji.

Read PDF

Similar papers

#machine learning Preprint Oct 2026

Energy-Aware Path Following: Comparative Analysis of Reinforcement Learning and NMPC for Electric Vehicles

Path-following control strategies typically follow the bi-objective optimization dilemma: minimizing deviations from a reference path while maintaining smooth speed profiles. The latter objective is especially relevant for Electric Vehicles (EVs), since their limited driving range can be extended by recovering energy t...

Mohamed Sabaa, M. Emam · 0 citations
Conference Sep 2026

ASSA-Based Nonlinear PID Parameter Tuning

As a typical underactuated and strongly coupled nonlinear system, the double-payload overhead crane (DPOC) introduces complex swing dynamics while improving material handling efficiency. Traditional parameter tuning methods struggle to balance position tracking accuracy, residual swing suppression, and control energy c...

Hua-Li Liu, Jun Wei, Xiao-Wei Su · 0 citations
Preprint Oct 2026

Risk-Constrained MPPI Tuning with Update-Aware MA-ES Racing

We formulate offline tuning of Model Predictive Path Integral (MPPI) controllers as a chance-constrained black-box optimization problem and develop an update-aware framework based on Matrix Adaptation Evolution Strategy (MA-ES). We use statistical information from noisy controller evaluations to determine when the evol...

M. Kimiaei, Roya Khalili Amirabadi, Mohsen Jalaeian-Farimani · 0 citations
Open access Oct 2026

Diagnosing the Sim-to-Real Gap in Latency-Aware Nonlinear Predictive Control: An Experimental Evaluation and Refinement Study on an Autonomous Vehicle

The study highlights how high-fidelity simulation can support controller development and diagnostic root cause isolation, while emphasizing the critical interplay between control design, state estimation fidelity, and hardware characteristics in Sim-to-Real deployment.

Nora Landaberea-Olazagoitia, Josu Pérez, G. Velez et al. · 0 citations
Open access Aug 2026

Multi-Trajectory Performance and Run-to-Run Consistency of GWO-Tuned PID Control for a Differential-Drive Mobile Robot

Accurate trajectory tracking of differential-drive mobile robots is strongly influenced by proportional–integral–derivative (PID) controller parameters, while a single fixed parameter set may exhibit different performance across trajectories with distinct geometric characteristics. This study investigates the multi-tra...

Auliya Nabila, Agung Muhamad Toha · 0 citations
Review Open access Sep 2026

Underwater-Mission-Oriented Objective Function Design for Metaheuristic Tuning of PID and Fractional-Order PID Controllers

Metaheuristic tuning of depth controllers for remotely operated vehicles is usually performed against a generic error integral, which does not encode mission priorities such as battery use, actuator stress from chattering, direction-dependent overshoot risk and post-event recovery. This paper treats the objective funct...

H. Ersoy, Aslihan Kartci · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.