Hierarchical Organization of Robots
The mapping from this list of 12 observable objects/events into a digit string is obvious: we use a 12-bit string and each bit takes the value 1 if the observation is made and 0 if it is not. The matrix S is also straightforward to devise. Action 1 is "collect and report", action 2 is "collect in bag for paper", action 3 is "collect in bag for cans", and so on. If one of the actions is "add to general garbage", another is "clean" and a third is "scrape from floor", then the actions column corresponding to the observation of gum will have three non-zero elements while the columns corresponding to "dirt" and "cigarettes" will have two.
A further issue here is the representation of the physical location of the robots. In many environments, the distance to be travelled to effect an action is obviously important. In this case, distances are not important so we simply represent the state digit string as having length equal to 12 times the number of rooms. The matrix S could repeat the same pattern of columns for the digit positions corresponding to each room or they could differ if something about each room requires a different set of maintenance actions. In the experiments reported here, we allowed for such differences. Each room has a supervisory robot to which four worker-robots report. With specified probabilities unique to each worker robot, there would be a malfunction of each of its action capacities at each task cycle. For some robots the probability of a specific malfunction was very low and for others quite high with a maximum probability of 0.4.
The independent executive robot noted at the end of each date which discrepancies between actual and target digit values were not being eliminated and, at the start of the following date, reorganized the worker robots so that the tasks not being effectively undertaken were reassigned to other robots working in the same room. No robot could have more that four tasks to perform.
Generated with CERN WebMaker