In this thesis, a Zeroth level Classifier System (ZCS) controller is compared to a hand-made controller (which features some simple rules to control the robot). The experiments are done in a world with continuous time and space. Their task is surveillance. The controllers control a LEGO Mindstorms robot with 4 wheels, a touch sensor, a pen device and a camera which can track the pen track. The pen track fades over time, and is used by the controllers to communicate indirectly.The experiments are done in three worlds. Each experiment features three robots with controllers of one type (either ZCS or hand-made). The ZCS controller has been optimized using a meta-evolutionary algorithm, but this did not yield better results than the not-optimized ZCS controller. Overall, the ZCS controller did not perform as good as the hand-made controller.
My master’s subject is Organisational Dynamics and Self Organization, an Artificial Intelligence master at the VU that focuses on the simple building blocks of systems. Those building blocks can be found everywhere: organic (think of the ants, which together constitute an anthill), organizations (all employees together are the company), economically (all transactions of traders determine the market price), etc.
My graduation project was internally (my supervisor was Martijn Schut) and focused on a specific evolutionary technology, “Zeroth level Classifier System (ZCS). This technique is used to control (LEGO) robots. The robots had to patrol an area (called a surveillance task). The robots, either a ZCS controller or a conventional controller, did compete in three different worlds.
Below is an example of an experiment in one of the three worlds. The robots have ZCS brains.
For those interested: My thesis and final presentation.