Abstract
<title>Abstract</title> <p> <bold>(1) Background:</bold> Training the next generation of agricultural practitioners demands hands-on, data-driven experience, yet field instruction is costly and hard to personalize; meanwhile the robots that could serve as tutors are battery-limited and must reason over uncertain, noisy field and learner data. <bold>(2) Methods:</bold> We present a genetic-fuzzy decision engine for energy-aware, personalized task-sequencing on agricultural education robots. A Mamdani fuzzy inference system (FIS) converts imprecise inputs — learner mastery, engagement, field suitability, and battery state — into soft per-task suitability and energy-caution scores; a genetic algorithm (GA) with order crossover, repair, and a fuzzy-weighted fitness evolves an energy-aware learning pathway subject to prerequisite, battery, and time constraints. The engine is embedded in a cloud–edge architecture with computation offloading, predictive energy modeling, missing-data imputation, and a standardized data layer, and is evaluated on real-data-grounded simulations with a simulated learner model. <bold>(3) Results:</bold> Against random, greedy, plain-GA, fuzzy-only, and Salp Swarm baselines over 30 independent trials, the genetic-fuzzy method attains the best mean fitness (1.346), significantly exceeding greedy, fuzzy-only, Salp Swarm, and random search (Friedman p = 9.4×10⁻²¹; pairwise Wilcoxon p < 10⁻³, rank-biserial r ≥ 0.88) and statistically matching the plain-GA ablation on clean data (p = 0.30). Its advantage emerges under uncertainty: as sensor data is randomly removed up to 50%, its fitness degrades by only 3.7% versus 11.2% for the plain GA — a gap that becomes statistically significant beyond 10% missingness (p < 4×10⁻³) — while computation offloading reduces energy per completed task by 42.1%. <bold>(4) Conclusions:</bold> Coupling fuzzy uncertainty handling with genetic optimization yields interpretable, energy-prudent personalized pathways whose principal benefit is graceful degradation under the noisy, missing-data conditions typical of real agricultural fields. </p>