COCOMO: Co-evolution of Control and Morphologies

The COCOMO project takes on the difficult challenge of automatically co-designing robotic bodies and brains, through the use of biology-inspired methods.


About the project

The project investigates how different types of control systems, building instructions, and building blocks affect the process of co-designing robotic bodies (morphologies) and brains (controllers). While parts of the EPEC project used real-world robots, the core methods of the COCOMO project are computer simulations of virtual robots. This allows us to investigate a much wider range of components in the design process. We also investigate the use of different types of algorithms to promote an exploratory search for new body-brain combinations, such as open-ended algorithms and quality-diversity approaches.


Exploiting Stepping Stones to Evolve Efficient Robot Morphologies

Crafting robots by hand can be both difficult and time consuming. While it is possible to manually design and control robots that work in predictable environments such as factories, the task becomes daunting when the robot is introduced to the diverse conditions that exist in the real world. We want to investigate how both the morphology and controller of a robot can be evolved automatically to tackle diverse environmental conditions. Although technology has not yet reached the point where a simulated robot can be automatically or easily manufactured and put to work in the real world, we would like to take a step towards making this possible. Our main research aim is to explore how efficient robot morphologies can be created automatically, and how robots can adapt to environments they face. We will also investigate how robots evolved in simulation can be applied in the real world, and whether adaptation to diverse environmental conditions in simulation can aid the robots in adapting to conditions faced in the real world.

Bio-inspired building blocks for robot design

In this sub-project, we investigate how different building blocks can be used to compose a robotic system. This concerns the components that constitute the robotic body, such as a range of modules with different capabilities, the types of controllers they can use, and how control is communicated across the different body parts. We also study different types of building instructions, called generative encodings, which can help generate more complex robot shapes.

Software outputs

Robotics, Evolution and Modularity (REM) 2D simulation framework

Selected publications

"Centralized and Decentralized Control in Modular Robots and Their Effect on Morphology"
by Mia-Katrin Kvalsund, Kyrre Glette, and Frank Veenstra.
To appear in ALIFE 2022.

Co-optimising Robot Morphology and Controller in a Simulated Open-Ended Environment
by Emma Hjellbrekke Stensby, Kai Olav Ellefsen, and Kyrre Glette.
In Applications of Evolutionary Computation. EvoApplications 2021, vol. 12694 of Lecture Notes in Computer Science, 2021, pp. 34-39. Details. Download: PDF.

MAP-Elites enables Powerful Stepping Stones and Diversity for Modular Robotics
by Jørgen Nordmoen, Frank Veenstra, Kai Olav Ellefsen, and Kyrre Glette.
Frontiers in Robotics and AI, vol. 8, 2021, Frontiers.

“How Different Encodings Affect Performance and Diversification when Evolving the Morphology and Control of 2D Virtual Creatures”
by Frank Veenstra and Kyrre Glette.
In Artificial Life Conference Proceedings, 2020, pp. 592-601, MIT Press.
Details. Download: PDF.

Published Jan. 29, 2022 6:33 PM - Last modified June 3, 2022 11:37 AM