Dynamically Optimizing Experiment Schedules of a Laboratory Robot System with Simulated Annealing

A schedulerhas been developed for an integrated laboratory robotsystem that operates in an always-on mode. The integrated system is designed for imaging plates containing protein crystallizationexperiments, and it allows crystallographers to enter plates at any time and request that they be imaged at multiple time points in the future. The schedulermust rearrange tasks within the time it takes to image one plate, trading off the quality of the schedulefor the speed of the computation. For this reason, the schedulerwas based on a simulated annealingalgorithm with an objective function that makes use of a linear programming solver. To optimize the scheduler, extensive computational simulations were performed involving a difficult but representative schedulingproblem. The simulations explore multiple configurations of the simulated annealingalgorithm, including both geometric and adaptive annealing schedules, 3 neighborhood functions, and 20 neighborhood diameters. An optimal configuration was found that produced the best results in less than 60 seconds, well within the window necessary to dynamically reschedule imaging tasks as new plates are entered into the system.