Continuous Alternation: The Complexity of Pursuit in Continuous Domains

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Stephen R. Tate, Professor and Department Head (Creator)
Institution
The University of North Carolina at Greensboro (UNCG )
Web Site: http://library.uncg.edu/

Abstract: Complexity theory has used a game-theoretic notion, namely alternation, to great advantage in modeling parallelism and in obtaining lower bounds. The usual definition of alternation requires that transitions be made in discrete steps. The study of differential games is a classic area of optimal control, where there is continuous interaction and alternation between the players. Differential games capture many aspects of control theory and optimal control over continuous domains. In this paper, we define a generalization of the notion of alternation which applies to differential games, and which we call "continuous alternation." This approach allows us to obtain the first known complexity-theoretic results for open problems in differential games and optimal control.

We concentrate our investigation on an important class of differential games, which we call polyhedral pursuit games. Pursuit games have application to many fundamental problems in autonomous robot control in the presence of an adversary. For example, this problem occurs in manufacturing environments with a single robot moving among a number of autonomous robots with unknown control programs, as well as in automatic automobile control, and collision control among aircraft and boats with unknown or adversary control.

We show that in a three-dimensional pursuit game where each player's velocity is bounded (but there is no bound on acceleration), the pursuit game decision problem is hard for exponential time. This lower bound is somewhat surprising due to the sparse nature of the problem: there are only two moving objects (the players), each with only three degrees of freedom. It is also the first provably intractable result for any robotic problem with complete information; previous intractability results have relied on complexity-theoretic assumptions.

Fortunately, we can counter our somewhat pessimistic lower bounds with polynomial time upper bounds for obtaining approximate solutions. In particular, we give polynomial time algorithms that approximately solve a very large class of pursuit games with arbitrarily small error. For e > 0, this algorithm finds a winning strategy for the evader provided that there is a winning strategy that always stays at least E distance from the pursuer and all obstacles. If the obstacles are described with n bits, then the algorithm runs in time (n/e)0(1), and applies to several types of pursuit games: either velocity or both acceleration and velocity may be bounded, and the bound may be of either the L2- or L&infin-norm. Our algorithms also generalize to when the obstacles have constant degree algebraic descriptions, and are allowed to have predictable movement.

Additional Information

Publication
Algorithmica, Vol. 10(2-4), 1993, pp. 151–181.
Language: English
Date: 1993
Keywords
Motion planning, Pursuit games, Differential games, Lower bounds

Email this document to