Assignment 2

Implement a genetic algorithm to evolve a simple creature that lives on a two-dimensional world and is able to move to the location of a mouse click as rapidly as possible. This creature should have a fixed-length bitstring genome that is made up of all of the creature’s behavioral rules. Each rule should have a certain number of conditional bits that encode the sensory conditions under which the rule is appropriate, and a certain number of action bits that encode what the creature should do when this rule is chosen. The creature must execute exactly one rule per time step. No rule may cause a creature to move more than one unit in X and one unit in Y per time step, nor to turn more than 90 degrees per time step. A creature may not both move and turn in the same time step, and must always move in the direction it is facing.

The evolution of the creature should start from a randomly generated initial population, and should demonstrate both crossing over and mutation. The full evolutionary process should take no longer than approximately 1-2 minutes on a consumer-grade PC. Please use the following Java source code, which has a grid world and GA framework already written for you.

http://www.ics.uci.edu/~wmt/courses/BiomoW04/GAFramework.zip

Note: In Eclipse, set Run->Run…Parameters to Width = 500, Height = 500 for best viewing.

There are five incomplete methods where it is necessary to add code:

Evolver.generateNewPopulation(); //please encode the process by which each successive generation is derived from the one that precedes it.

Creature.sense();//please encode the process by which the creature’s sensory apparatus detects the target (e.g., the mouse click).

Creature.chooseAction();//please encode the process by which a rule is chosen based on how well its conditional bits match what it sensed in Creature.sense().

Creature.takeAction();//once a rule has been chosen, this method shows how that rule is interpreted to affect the position or forward vector of the creature.

TestWorld.testFitness(); //this method determines the fitness function by which the creatures are evaluated.

In addition to the above methods, it will probably be necessary for you to add other methods and fields to the code in order for it to work well. Please comment all code that you add so that it is clear what each section is doing.

Please email wmt@uci.edu if you have any questions, and email an entire set of compiling files to wmt@uci.edu by Monday, 2/2/04 at 5pm.

Note: If you’d like a bigger challenge, see how efficiently you can evolve an optimal creature, minimizing number of rules, number of bits per rule and total processing time.