Brent E. Eskridge

Imitating success: a memetic crossover operator for genetic programming

For some problem domains, the evaluation of individuals is significantly more expensive than the other steps in the evolutionary process. Minimizing these evaluations is vital if we want to make genetic programming a viable strategy. In order to minimize the required evaluations, we need to maximize the amount learned from each evaluation. To accomplish this, we introduce a new crossover operator for genetic programming, memetic crossover that allows individuals to imitate the observed success of others. An individual that has done poorly in some parts of the problem may then imitate an individual that did well on those same parts. This results in an intelligent search of the feature-space, and therefore fewer evaluations.

Memetic Crossover for Genetic Programming: Evolution Through Imitation

For problems where the evaluation of an individual is the dominant factor in the total computation time of the evolutionary process, minimizing the number of evaluations becomes critical. This paper introduces a new crossover operator for genetic programming, memetic crossover, that reduces the number of evaluations required to find an ideal solution. Memetic crossover selects individuals and crossover points by evaluating the observed strengths and weaknesses within areas of the problem. An individual that has done poorly in some parts of the problem may then imitate an individual that did well on those same parts. This results in an intelligent search of the feature-space and, therefore, fewer evaluations.

Effects of Local Communication and Topology on Collective Movement Initiation

Collective movement in autonomous systems, such as a team of robots, are frequently implemented using complex interaction rules and have significant communication requirements. These restrictions frequently relegate such systems to static, simplified environments. In contrast, collective movements in natural systems consistently occur in dynamic, complex environments in which significant communication is either impractical or impossible, and have been successfully modeled using simple, local interaction rules. In the work presented here, one such model is extended to include local communication and the spatial distribution of the group so that it can eventually be used as a guide for developing artificial systems capable of cohesive, collective movements. The extended model predicts that a reliance on local communication does not necessarily mean there will be a significant loss in the expected success of collective movement attempts if appropriate interaction rules are chosen. Furthermore, the model predicts that the addition of local communication, in conjunction with the topology of the group, results in higher expected success in attempting collective movements for individuals with central locations in the group as compared to individuals occupying edge locations.

Extending adaptive fuzzy behavior hierarchies to multiple levels of composite behaviors

We propose an extended version of adaptive fuzzy behavior hierarchies, termed Multiple Composite Levels (MCL), that allows for the proper modulation of composite behaviors over multiple levels of a behavior hierarchy, and demonstrate its effectiveness for a hybrid learning/reactive control system. Controllers using adaptive fuzzy behavior hierarchies have previously been shown to provide effective control for robots tasked with multiple concurrent tasks. However, when more complex hierarchies are used to provide control for tasks of increasing complexity, low-level reactive behaviors may not be properly weighted, resulting in sub-optimal control. Through experimental evaluation in which composite behaviors that coordinate lower behaviors are learned using reinforcement learning, we demonstrate that MCL provides effective control in a complex multi-agent task, whereas the original implementation of adaptive fuzzy behavior hierarchies does not.

Nurturing promotes the evolution of learning in uncertain environments

Adapting to a changing and uncertain environment is vital for the long-term success of individuals, whether they are biological or artificial. While learning can be powerful in the adaptation process, a lack of understanding exists in the factors that promote or inhibit its evolution. Nurturing is widely thought to be a contributing factor, if not a requirement, for high levels of learning. In the study described here, we investigated how nurturing contributes to the evolution of learning in uncertain environments using a simple, biologically inspired evolutionary simulation. The simulation predicts that the two models of nurturing used here, nurturing as social learning and nurturing as a means of safe exploration, both promote the evolution of learning in uncertain environments in which learning would otherwise not be a viable strategy. The results also indicate that nurturing is also a factor in improving fitness in environments in which learning is already viable.

Is "best-so-far" a good algorithmic performance metric?

In evolutionary computation, experimental results are commonly analyzed using an algorithmic performance metric called best-so-far. While best-so-far can be a useful metric, its use is particularly susceptible to three pitfalls: a failure to establish a baseline for comparison, a failure to perform significance testing, and an insufficient sample size. The nature of best-so-far means that it is highly susceptible to these pitfalls. If these pitfalls are not avoided, the use of the best-so-far metric can lead to confusion at best and misleading results at worst. We detail how the use of multiple experimental runs, random search as a baseline, and significance testing can help researchers avoid these common pitfalls. Furthermore, we demonstrate how best-so-far can be an effective algorithmic performance metric if these guidelines are followed.

Using priorities to simplify behavior coordination

Previous research has used behavior hierarchies to address the problem of coordinating large numbers of behaviors. However, behavior hierarchies scale poorly since they require the state information of low-level behaviors. Abstracting this state information into priorities has recently been introduced to resolve this problem. In this work, we evaluate both the quality of priority-based behavior hierarchies and their ease of development. This is done by using grammatical evolution to learn how to coordinate low-level behaviors to accomplish a task. We show that not only do priority-based behavior hierarchies perform just as well as standard hierarchies but that they promote faster learning of solutions that are better suited as components in larger hierarchies.

Using action abstraction to evolve effective controllers

We propose that abstracting the actions of a behavior coordination mechanism promotes the faster development and higher fitness of an effective controller for complex, composite tasks. Various techniques are well suited for the development of controllers for individual simple tasks. However, as individual tasks are combined into complex, composite tasks, many of these techniques quickly become impractical. By reusing existing behaviors, the focus of development for a controller can be shifted from low-level control to high-level coordination of these existing behaviors. As a result, the development of an effective controller becomes far more practical. Experiments using a single-agent task in a continuous environment demonstrate that grammatical evolution is capable of discovering fuzzy rulesets which effectively coordinate existing behaviors in a controller in fewer generations and with higher fitness than monolithic controllers.

Prioritizing Fuzzy Behaviors in Multi-robot Pursuit Teams

The combination of fuzzy control and behavior hierarchies allows for the construction of more complex behavior-based robot control agents than does either technique alone. However, current implementations are limited in their complexity since high-level behaviors still use low-level sensor information. We propose a technique for abstracting this low-level sensor information into priorities which are used to completely abstract out the context in which a high-level, fuzzy behavior operates. This modification enables a single high-level behavior to coordinate the lower-level behaviors within a single robot, among the robots in a team, or even among teams of teams. This is demonstrated in a scenario in which multiple pursuers attempt to capture a prey.

Emergence of Leadership within a Homogeneous Group

Large scale coordination without dominant, consistent leadership is frequent in nature. How individuals emerge from within the group as leaders, however transitory this position may be, has become an increasingly common question asked. This question is further complicated by the fact that in many of these aggregations, differences between individuals are minor and the group is largely considered to be homogeneous. In the simulations presented here, we investigate the emergence of leadership in the extreme situation in which all individuals are initially identical. Using a mathematical model developed using observations of natural systems, we show that the addition of a simple concept of leadership tendencies which is inspired by observations of natural systems and is affected by experience can produce distinct leaders and followers using a nonlinear feedback loop. Most importantly, our results show that small differences in experience can promote the rapid emergence of stable roles for leaders and followers. Our findings have implications for our understanding of adaptive behaviors in initially homogeneous groups, the role experience can play in shaping leadership tendencies, and the use of self-assessment in adapting behavior and, ultimately, self-role-assignment.