| On 2007-08-27 at 16:30:00 (Brussels Time) |
Abstract
Self-replication to date has only been demonstrated in a planned manner or via cellular automata or as deterministic assembly of machines, that too in a scarce fashion and hence lacks significant thrust. In this paper I propose to cover this void by adapting a novel approach of growing a neural network topology to control the modular robot morphology and attaining full edged self replication by the process of fusion and fission in a simulated world of sophisticated physics. Modular reconfigurable robotic systems that are composed of many modules have three promises, to be versatile, robust, and low cost. I have developed a new model, Reinforced Central Pattern Generated Focused Time Delay Neural Network and deployed over stable and dock-able cylindrical modular morphology. It also develops novel manoeuvring capabilities directed towards self-replication. The vision is to auto-colonise vacant parts of space using these exponentially self-replicating creatures. I also ta ke a wider perspective of this problem in the robotics domain introducing various theories of state estimation in a real time control scenario.
Keywords
Robotics, Self-replication
