Autonomous Heterogeneous Mining Fleet Control at Nonorthogonal Intersections by Hippopotamus Optimization Algorithm-Based Adaptive Proximal Policy Optimization

Abstract

This work focuses on the autonomous heterogeneous fleet cooperative control problem in nonorthogonal intersections and proposes a fleet control algorithm to give the fleet target acceleration to achieve longitudinal control of the fleet with the objective of minimizing travel delay. We propose a hybrid hippopotamus optimization algorithm (HOA)-based adaptive proximal policy optimization (PPO) algorithm that leverages HOA for adaptive learning rate tuning in PPO. It addresses the challenge of controlling the operation of a heterogeneous fleet of trucks by calculating the priorities of all trucks in the intersection area and incorporating them into the state space of the reinforcement learning. Furthermore, our framework’s state-action space design is universally adaptable, functioning robustly across all intersection types, including those with non-orthogonal angles. Finally, the simulation experiments demonstrate that our proposed method can be adapted to different operational scenarios and obtain better policy with fewer training episodes than traditional learning rate setting methods.