Abstract: | The machine produced in this work provides information about how far a given object is from its goal, given that there is no knowledge of what the objects are in the state or how they move within that state. The information provided is merely a description of the start state together with the states that follow on from that state. The machine has the ability to identify an object (king) and learn its moves on an empty board. It can then represent those moves in a state of arbitrary size. The distance that the king is from a given goal element has also been learnt for an arbitrarily sized board. These concepts are represented in terms of rooted polynomial functions and can be used as heuristics providing that the location of the goal is known. This machine has not yet been fully tested on a whole selection of domains.
|