In the paper titled “Merging Pruning and Neuroevolution: towards Robust and Efficient Controllers for Modular Soft Robots” (cambridge.org) the researchers employ an evolutionary algorithm, a form of computation loosely inspired by natural evolution, for optimising the controller of Voxel-based Soft Robots. These simulated robots are a kind of modular, biologically inspired soft robots that can be controlled using artificial neural networks. While evolving the robots to be fast at locomotion, the researchers apply pruning to the networks constituting the robot brains. Pruning is the process of removing neurons or connections; pruning reduces the complexity, increases the energy efficiency and in some cases improves the generalisation capability.
Through their work they found that “incorporating some forms of pruning in neuroevolution leads to almost equally effective controllers as those evolved without pruning, with the benefit of higher robustness to pruning”. They also observed occasional improvements in generalisation ability.
The work is based on a collaboration between University of Trieste, OsloMet AI Lab and NordSTAR, and led by Giorgia Nadizar. Nadizar is a former research assistant at OsloMet and current PhD student at the University of Trieste through the BioSoftRob project (nichele.eu).
The paper is published in the Knowledge Engineering Review (level 2 journal in Norway) by Cambridge University Press.
In the top photo you can see the involved researchers from the University of Trieste. From left: Marco Zullich, Felice Andrea Pellegrino, Giorgia Nadizar and Eric Medvet. Photographer: Erica Salvato.