Planning under Uncertainty with Weighted State Scenarios

The state description and corresponding dynamic Bayesian network are shown in Table 1 and Figure 1. In the paper we mentioned that ms denotes the state of task i. To make this more detailed, we factorms into a tuple (m i r,m i d,m i s,m i c). The variable mr encodes the release time of task i. The variable md represents the number of steps task i can still be postponed, which can be used to encode the deadline. The variable my represents the number of steps task i still has to run, which can be used to encode the length of the task. The last decision of agent i is represented by mc, which is an auxiliary variable which we need to reduce the size of the action space. This variable can be either run (R) or idle (I). The factored state variables can be used to determine which actions are feasible to execute given the current state, which we discuss below.