Integrated Reinforcement and Evolutionary Learning Algorithm: Application to Alife in a Shared World

Abstract

The reinforcement learning algorithm we have developed for simulating Alife in a shared environment has proved to be quite effective in implementing an easy adaptation of an organism to an environment at an individual level while the evolutionary learning algorithm is effective in evolving the capacity to internally generate and develop its own teaching input to guide learning at a population level. The reinforcement signal reflecting interactions with changes in environment such as the number of food elements and their distribution facilitates the task of organisms’ finding and then eating food efficiently in a shared environment at an individual level while the evolutionary process produces a useful feedback from the environment updating the teaching input. The integration of the reinforcement signal and the evolutionary process shows an excellent improvement in performance of food eating capability showing the significance of the integrated learning.