Lamarck’s Revenge: Effects of Lamarckism on Morphologically Evolvable Robots

Evolutionary robot systems offer two principal advantages: an advanced way of optimising robots through evolution and a special research platform to conduct what-if experiments regarding questions about evolution. Our study sits in the intersection of these. We investigate the question ``What if the 18th century biologist Lamarck was not completely wrong and individual traits acquired during lifetime could be passed on to offspring through inheritance?'' Specifically, we implement an evolutionary robot system where morphologies (bodies) and controllers (brains) are both evolvable and robots also can improve their controllers through learning. Within this system, we compare a Lamarckian method, where learned bits of the brain are inheritable with a Darwinian method, where they are not. 

We test both methods in two tasks, Point navigation and Panoramic rotation, and evaluate them based on three criteria: robot performance, morphology, and behaviour. Our experiments aim to answer key research questions about the efficiency of these frameworks, their effect on the evolved robot morphologies, and the resulting robot behaviors. The results demonstrate that Lamarckian evolution is not only viable in an artificial evolutionary system but also significantly outperforms its Darwinian counterpart.