Action-Based Discretization for AI Search

2754 Words Oct 13th, 2012 12 Pages
Action-Based Discretization for AI Search
Dr. Todd W. Neller*
Department of Computer Science Gettysburg College Campus Box 402 Gettysburg, PA 17325-1486 Introduction As computer gaming reaches ever-greater heights in realism, we can expect the complexity of simulated dynamics to reach further as well. To populate such gaming environments with agents that behave intelligently, there must be some means of reasoning about the consequences of agent actions. Such ability to seek out the ramifications of various possible action sequences, commonly called “lookahead”, is found in programs that play chess, but there are special challenges that face game programmers who wish to apply AI search techniques to complex continuous dynamical systems.
…show more content…
Finally, we compare three different means of action-parameter discretization, including a dispersion technique that generates an approximate uniform sampling of closed action-spaces. Action-Based Discretization Artificial Intelligence search algorithms search discrete systems, yet we live and reason in a continuous world. Continuous systems must first be discretized, i.e. approximated as discrete systems, to apply such algorithms. There are two common ways that continuous search problems are discretized: state-based discretization and action-based discretization. Statebased discretization becomes infeasible when the state space is highly dimensional. Actionbased discretization becomes infeasible when there are too many degrees of freedom. Interestingly, biological high-degree-of-freedom systems are often governed by a much smaller collection of motion primitives [Mataric, 2000]. We focus here on action-based discretization. Action-based discretization consists of two parts: (1) action parameter discretization and (2) action timing discretization, i.e. how and when to act. See Figure 1. The most popular form of discretization is uniform discretization. It is common to sample possible actions and action timings at fixed intervals. For the following algorithms, we focus on action-timing discretization. Experimental evidence of this paper and previous studies [Neller, 2000] suggests that a fixed
Open Document