@inproceedings {pub6463,
	title = {A Learning Large Neighborhood Search for the Electric Autonomous Dial-A-Ride Problem},
	author = {Laurenz Tomandl AND Maria Bresich AND Guenther Raidl AND Steffen Limmer},
	year = {2026},
	month = {April},
	abstract = {In this work, we explore improvements to the Electric Autonomous Dial-A-Ride Problem by introducing a Learning Large Neighborhood Search (LLNS) strategy. In contrast to previous studies that focused on sparsification or hyperparameter tuning via deep learning, our approach learns a function that evaluates how promising a route is for inclusion in the destroy set of the LNS. Specifically, we employ a route-wise destroy operation that removes entire routes from the solution. Each route is scored by a neural network and based on this information selected to be destroyed in a probabilistic manner. The neural network model evaluates routes using features such as route duration, the number of covered orders,average user detour, energy consumption margin, time-window slack, and how long the route was already part of the incumbent solution, and when it has been selected the last time for destroy. We employ a reinforcement learning strategy for training the neural network offline on representative problem instances. The reward is based on the improvement in objective value after removing a route and repairing the partial solution.},
	publisher = {Springer, LNCS, LNAI, LNBI},
	booktitle = {International Conference on Computer Aided Systems Theory (Eurocast) 2026}
}