Dynamic Task Allocation for Efficient Container Handling at ECT Delta Terminal

Abstract

This thesis addresses the problem of Automated Guided Vehicle (AGV) scheduling by employing a Multi-Agent pickup and delivery problem (MAPD) formulation, which we identified as a good approach compared to the other approaches considered in this paper for tackling the related routing and scheduling sub-problems inherent to this domain. We adjusted a simulation environment to experiment with different schedules and routes, addressing critical factors such as collisions, reserved crane parking, and the effects of randomness and variability. To optimize AGV operations, we utilized three key heuristics: idle time, collision time, and drive time. For routing, we implemented cooperative A* with deadlock prevention, ensuring robust, collision-free task completion.We introduced a new initialization method, the greedy method, which uses simulation results to generate a starting solution. Our approach explored the solution space from this starting solution using local search algorithms, specifically random, swap, and insertion operators combined with activity time blocks. We enhanced the search process with Iterated Local Search (ILS), employing iterated greedy neighbors (IG) to improve the solution space exploration, outperforming random neighbors (RN) by maintaining beneficial solution characteristics. Periodic rescheduling was implemented to manage the randomness of crane times, with experimental results confirming its necessity for improving solution robustness. Overall, our combined heuristics and search strategies effectively minimized the total container handling time at terminals.