AI Planning for Destination Control in Elevators

In a destination control system, passenger enter their desired destination floor via some input device and are subsequently allocated to a specific elevator in a group of elevators. Since the destination is known in advance, new opportunities to optimize the travel routes of elevators become available. Additionally, individual service requirements for each passenger can be defined:

The control software combines various AI techniques as the basis of a highly optimized and scalable search algorithm that is able to search and prune search spaces of up to $10^{10}$ states. In addition, the algorithm can handle the constraints described above. The AI-based controller was embedded into a service-oriented architecture that coordinates terminals dispatching calls with a group of elevators using agent and auction technology. The developed solution permitted the commercial breakthrough of destination control systems in the elevator industry and became the general foundation for many Schindler elevator products because of its performance, robustness, easy configuration, and general applicability to all variants of elevator control systems.

Schindler maintains various pages about the technology that is part of various product offerings. Here are some links (if the pages have moved, just search for "Schindler destination control"):

The AI-based control algorithm has enabled the breakthrough of destination control. It is the most sophisticated algorithm in use within the elevator industry, yields up to 30 % higher performance, and received several prestigious awards, e.g. the Breaking Barriers Award of the European Commission for its better service to disabled passengers when compared to conventional control systems.



A model of this domain in PDDL was used in the final round of the International Planning Competition in 2000. Variants of miconic10 domains have been used in subsequent competitions, see the official competition webpage.

Jana Koehler, 10/August/2014