This project consists of a re-implementation and novel interpretation of Karl Sims’ 1994 paper titled “Evolving Virtual Creatures”. The paper presents a method for generating, evaluating, and mutating virtual 3D genotypes, which results in a remarkable diversity of solutions to various tasks. We aim to recreate and expand upon the success of the original publication by using modern engines, asynchronous cloud training, and reinforcement learning theory, focusing on optimizing for 3D swimming. Developers will comprehend academic papers, implement algorithms from the fields of evolutionary computation and reinforcement learning, and build a functioning interface in Unity.
We are looking for developers who love machine learning and have experience in C#, Python, or C. Experience/interest in reinforcement learning, Unity, and/or evolutionary algorithms is also a plus. Email firstname.lastname@example.org, or DM “Andrew Magnuson” in the UTMIST Discord if you’re interested!