Evolving integrated low-level behaviors into intelligently interactive simulated forces

Abstract

Combining optimal high-level planning with low-level behaviors has usually been accomplished through two separate mechanisms. Typically, both of these mechanisms have relied upon heuristic approaches to control the behaviors of simulated agents to achieve mission goals. Recent research into evolving optimal high-level tactical behaviors for simulated vehicles proved quite fruitful even when heuristics were utilized to navigate low-level terrain. Evolutionary programming was used to optimally control computer generated forces (CGFs) on two opposing teams in highly dynamic environments. Tactical courses of action were learned adaptively for individual vehicles as well as for higher-level aggregations (i.e., platoons). Evolutionary updates of behavioral plans incorporated dynamic changes in the developing situation and the sensed environment.

Inconsistencies can, however, arise when low-level heuristic navigation systems produce actions that significantly deviate from the evolved high-level tactical plans. Estimation and prediction of behaviors can result in erroneous outcomes due to the lack of integration between the planning mechanisms. This paper presents a method for solving these problems by using evolutionary programming for both high-level tactical planning and low-level navigation through arbitrary terrain. The resulting software, which builds upon previous work, uses no heuristics in the planning process. The evolved behaviors more accurately reflect the realities of planning in the face of nonhomogeneous terrain, and present a much improved simulation tool for training.