@inproceedings {pub6461,
	title = {Optimizing Dynamic Dispatching in Elevator Control with a Destination Registration System},
	author = {Maria Bresich AND Guenther Raidl AND Steffen Limmer},
	year = {2026},
	month = {April},
	abstract = {In this work, we tackle the assignment and dispatching problem for an elevator group control system with a destination registration system (EGCS-DRS). Our considered system also allows for specifying the group size in case multiple people arrive together which further increases the practical complexity of this NP-complete problem. The goal is to minimize the average waiting time from requesting an elevator until entering it.
We first address the static (offline) problem where all orders are known in advance and formalize it as a MILP before approaching the dynamic (online) variant in the same way. As such exact approaches often struggle to scale to larger instances, we also propose a fast greedy algorithm based on a simple heuristic for the dynamic problem.  At each decision point, the algorithm iteratively determines only the next best action for each elevator which includes whether to move or not and which passengers to pick up next. A similar approach is used in our third proposed method for the online problem, a reinforcement learning algorithm with the goal of predicting for each elevator the best next destination. Our approaches are evaluated in a computational study on different elevator system sizes, two scenario lengths, and various traffic patterns inspired by real-world use cases. Results show that the offline MILP approach is only feasible for small instances whereas the online MILP provides high quality solutions also for larger instances but requires too much time to be used in real-time applications. In terms of computation time, the greedy algorithm is most efficient while also achieving good solutions for most instances. The performance of the RL approach varies across the different traffic patterns and elevator systems but it is in general also able to find good solutions even for large instances.
},
	publisher = {Springer, LNCS, LNAI, LNBI},
	booktitle = {International Conference on Computer Aided Systems Theory (Eurocast) 2026}
}