“…Apart from sampling and calculating the robust time costs in order to determine the RPI and IPI, the procedural details for the resulting PI-candidate algorithms are almost exactly the same as baseline PI, see [3,8,26] and [38]. However, there is another important difference between the baseline and the PI-candidate architecture; when using a candidate module, each agent i communicates its own estimate of the relevant, local uncertain parameters to the others (these include estimates of its own velocity Θ i and location Υ x,i , Υ y,i , and Υ z,i , and its estimates of all the task locations x,i , y,i , and z,i , and durations D i ).…”