Elevators seem straightforward—press a button, wait for arrival—but their dispatching algorithms involve complex optimization challenges that affect how long riders actually wait.
The simplest elevator algorithm, SCAN, patented in 1961, travels from the lobby to the top floor before reversing direction, picking up and dropping off passengers along the way. A refinement called LOOK stops only at requested floors before reversing, which most people expect. These basic algorithms work fine for single-car systems, but multiple elevators require coordination.
When multiple cars operate in the same bank, a central scheduler assigns requests to elevators. The naive approach assigns each call to the nearest car, but according to the source material, this can be improved. Otis’ Relative System Response (RSR) algorithm scores each car’s suitability by considering factors like anti-bunching (avoiding multiple cars heading to the same floor) and rewarding idle cars within two floors of the caller. RSR also re-optimizes every 5 seconds, allowing the system to reroute passengers between elevators if delays occur.
Measuring elevator performance goes beyond simple averages. The source material explains that riders fixate on worst-case scenarios rather than average wait times. Metrics like p90 (90th percentile) reveal more meaningful user experience data—a p90 of 2 minutes means 90% of rides involve a 2-minute or shorter wait. Traffic patterns vary significantly by time of day. Morning rush in office buildings creates notoriously poor wait statistics, as nearly all traffic flows from lobby to upper floors.
Benchmarking LOOK against RSR reveals surprising results. At higher flow rates when elevators are consistently full, LOOK actually outperforms RSR, as the extra optimization rules provide diminishing returns. LOOK also performs better in smaller buildings with fewer elevators per bank, suggesting simpler algorithms often suffice.
Countintuitively, newer destination dispatch systems—which use kiosks to pre-assign passengers to specific elevators—generally produce worse wait times than traditional up-and-down buttons. The kiosk’s rigidity prevents the system from adapting to changing conditions, whereas the traditional button system allows constant re-optimization. As the source notes, the loss of flexibility outweighs the benefit of knowing passenger destinations in advance.
Key facts
- The SCAN algorithm, patented in 1961, represents the simplest elevator logic, traveling fully to top before reversing
- The LOOK algorithm stops only at requested floors, matching most people’s expectations
- Otis’ RSR algorithm optimizes elevator assignments every 5 seconds and can reroute passengers mid-journey
- Morning rush creates the worst elevator wait statistics in office buildings due to one-directional traffic flow
- Destination dispatch kiosks typically perform worse than traditional buttons because they prevent real-time optimization
- Simple LOOK algorithm can outperform more complex RSR at high flow rates and in small buildings