Animats and Genetic Algorithms for Orientation

Orientation theory is not just a conceptual framework for understanding system evolution and behavior under the exergy availability constraint. It also allows quantitative and comparative analysis of system performance under different environmental conditions.

Genetic algorithms are models of biological adaptive processes that are being widely and successfully applied to a wide spectrum of adaptation and optimization problems. In particular, these algorithms have been used to simulate learning and adaptation of artificial animals (ani-mats) in simulated environments containing food and obstacles. They can be used to demonstrate the emergence of basic orientors in self-organizing systems having to cope with complex environments.

The animat model incorporates essential features of a simple animal in a diverse environment. Being an open system, an animal depends on a flow of exergy from the environment. In the course of its (species) evolution, it has to learn to associate certain signals from the environment with reward or pain and to either seek or avoid their respective sources (exergy gain or exergy loss). This learning phase (of populations) will eventually lead to the establishment of cognitive structure and behavioral rules which are approximately optimal in the particular environment (with respect to maximization of reward, minimization of pain, and securing survival). These behavioral rules incorporate knowledge which enables intelligent behavior.

The animat is designed to simulate this process. It can pick up sensory signals from its environment (containing food and obstacles), and classify them with available rules to determine an appropriate action (direction of movement). After a successful move, the strength of rules leading up to it is increased by sharing in the reward (i.e., exergy gain). New rules are occasionally generated by either random creation, or by genetic operations (crossing-over and recombination). They are added to the existing rule set, and compete with the other rules for reward. Unsuccessful rules are not reinforced and lose strength and influence in the rule set.

The training process consists of placing the animat at a random empty location in an environment with specific environmental properties, and allowing it to move around searching for food. A collision with an obstacle causes a loss of exergy and throws the animat back to its previous position. Rules leading to success are rewarded. A genetic event of rule generation may occur with a prescribed probability. Random rules are created in unknown situations. The process is repeated for a large number of steps (typically 10 000). Eventually, a set of behavioral rules develops which allows optimal behavior under the given set of conditions.

Note that this optimal behavior has not been defined in terms of an objective function guiding the evolution of the set of behavioral rules. The rule set develops solely from the reinforcement of rules which lead to food or avoid collisions. An explicit exergy balance accounts for all exergy losses associated with movement, collisions with obstacles, and rule generation, and exergy gains due to uptake of food. The development of the rule set is then driven by the requirement to optimize exergy pickup in the given environment (with specific resource availability), while allowing for environmental variety, variability, and change specific for that environment. Neglect of these properties is penalized by lack of fitness, and threat to survival, and causes disappearance of deficient rules. Other criteria besides efficiency will therefore be reflected in the set of behavioral rules. Since these were not expressly introduced, we must recognize them as emergent value orientations or objective functions.

The animat experiment contains all components necessary for a study in the basic orientor framework. Animat fitness depends on the ability to maintain a positive exergy balance in the long term. This exergy balance is therefore at the core of the orientor satisfaction assessment. At each step, exergy uptake (by food consumption) and exergy losses (by collisions with obstacles, motion, and learning of rules) are recorded and used to compute the momentary exergy balance. Attention to all orientors is mandatory to ensure a positive exergy balance even under adverse environmental conditions.

Quantitative measures must be defined for characterizing the different properties of the environments used in the animat experiments. Animat performance in different environments is compared by using measures of orientor satisfaction. These have to be defined using relevant parameters of animat performance.

Effectiveness

Effectiveness

Adaptability

Figure 4 In an identical training environment, different lifestyles may evolve. Generalists stress freedom of action, specialists focus on effectiveness, while cautious types emphasize security.

Adaptability

Generalist (F) -«- Specialist (E) —- Cautious (S)

Figure 4 In an identical training environment, different lifestyles may evolve. Generalists stress freedom of action, specialists focus on effectiveness, while cautious types emphasize security.

Was this article helpful?

0 0
Oplan Termites

Oplan Termites

You Might Start Missing Your Termites After Kickin'em Out. After All, They Have Been Your Roommates For Quite A While. Enraged With How The Termites Have Eaten Up Your Antique Furniture? Can't Wait To Have Them Exterminated Completely From The Face Of The Earth? Fret Not. We Will Tell You How To Get Rid Of Them From Your House At Least. If Not From The Face The Earth.

Get My Free Ebook


Post a comment