Recently, the electrification and automation of heavy-duty trucks has gained significant attention from both industry and academia, driven by new legislation introduced by the European Union. During a typical drive cycle, the mass of an urban service truck can vary substantially as waste is collected, yet most existing studies rely on a single controller with fixed gains. This limits the ability to adapt to mass changes and results in suboptimal energy usage. Within the framework of the EU-funded OBELICS and ESCALATE projects, this study proposes a novel control strategy for a semi-autonomous refuse truck. The approach combines a particle swarm optimization algorithm to determine optimal controller gains and a multiple model controller to adapt these gains dynamically based on real-time vehicle mass. The main objectives of the proposed method are to (i) optimize controller parameters, (ii) reduce overall energy consumption, and (iii) minimize speed tracking error. A cost function addressing these objectives is formulated for both autonomous and manual driving modes. The strategy is evaluated using a real-world drive cycle from Eskişehir City, Turkiye. Simulation results show that the proposed MMC-based method improves vehicle performance by $5.19\%$ in autonomous mode and $0.534\%$ in manual mode compared to traditional fixed-gain approaches.