Evolutionary robotics has yet to achieve mainstream status within the robotics research community, and among the reasons for this is the relative low maturity of the field. Most of the current research is still done in the scope of fundamental research, and has yet to be used to solve a wide range of real world problems, although much of it shows great promise. Conventionally designed and controlled robots solve an ever increasing number of tasks. For evolutionary robotics to catch up with, or even outperform traditional robotics techniques, a wide array of functional real life robots built on the foundation of evolutionary techniques is required.
This thesis proposes an evolutionary framework for evolving both morphology and control for a six legged robot. It features a parameterized 3D model with adaptable servo placement, base size, leg lengths, and a possibility for adding two more legs or tool holders to the front. It is also integrated into a simulation environment for evolutionary experiments. The algorithm tested is able to produce a varied set of solutions with different weights and speeds, and shows promise for solving more complex tasks or fitness functions.
The thesis also tests whether co-evolution of control and morphology is a feasible technique for robot design, by comparing the performance of a manually designed instance of the robot to two evolved models, both in simulation and reality. Machine learning is also used to lessen the reality gap present between the simulations and the physical experiments. Co-evolution of control and morphology shows a significant improvement over the manually designed morphology and gait, producing a robot which is 3% lighter and 49% quicker.