Abstract:

We introduce a novel group elevator scheduler based on an explicit decision-theoretic calculation for the expectation of any polynomial function of passengers' waiting times, marginalized over sources of uncertainty in the system state. The same framework can be used to identify risky assignments that have low expected costs but may cause some individuals to be "stranded" with excessive waits for service. The resulting scheduler avoids risky assignments while minimizing average passenger waits.Experimental results demonstrate that the method can reduce the variance of waiting times and the fraction of passengers wiating excessively in comparison with risk-neutral schedulers, while still achieving better waiting times than reference ETA controllers.