Skip to content. | Skip to navigation

Personal tools
You are here: Home Theses Modeling Evolutionary Algorithm Optimized Autonomous Sensory Agents in ROS


Prof. Dr. S. Decker
RWTH Aachen
Informatik 5
Ahornstr. 55
D-52056 Aachen
Tel +49/241/8021501
Fax +49/241/8022321

How to find us

Annual Reports





Modeling Evolutionary Algorithm Optimized Autonomous Sensory Agents in ROS

Thesis type
  • Master
Student Mohammad Touhidur Rahman
Status Finished
Submitted in 2018
File download

In the context of Phoenix project, one current challenge is to add more physical properties to the Evolutionary Algorithm loop, in addition to, simulating more complex communication schemes. Robot operating system (ROS) is a very powerful middleware that is widely used in the robotic community in similar cases as ours. It facilitates parallel behavior simulations and many physical properties, such as weight, which will help in the EA evaluation part.

Evolutionary robotics is a field that applies evolutionary algorithms to problems related to both robot control and morphology. More than 20 years ago, researchers started exploring the potential of the evolutionary algorithm in the field of robotics. Evolving robot morphology is an attractive yet complex area of research. Past and recent works on it showed promise in various areas like generating evolved structures for locomotive robots, reducing the reality gap between simulated and real-world evolution, achieving more morphological complexities and so on. Humans can build a robot structure to perform various tasks, but the performance improves with modification to the structure over a long period of time and testing. Morphological evolution techniques can alleviate this problem. Robots can learn and evolve in the evolutionary cycle to generate a structure that will be more adapted to the environment and tends to perform better towards achieving the tasks.

We build a simulation framework to perform a morphological evolution of virtual robotic agents. We use ROS (the robot operating system), Gazebo and state of the art UUV simulator tool to build it. The agents evolve to explore environments that are difficult to access. The framework enables the simulation of such environments. We focus on one aspect of the morphology, i.e. the mass of an agent for evolution. We divide the agent in a grid-like structure to provide the mass to the agent. This structure evolves in the evolutionary cycle to perform in the simulated environment. We execute four experiments in four different environments. The experimental results prove the frameworks ability to evolve the agents in simulated environments.

Document Actions