What is it about?
One way of optimizing at the same time the body of the brain of modular robots is through a two-levels optimization scheme: in the inner loop, you let the brain learn how to control the body; in the outer, you search in the space of bodies through an evolutionary alforithm. This paper shows that the encoding for the body in the outer loop does impact on the learning ability, performed through reinforcement learning, of the inner loop.
Featured Image
Photo by Robo Wunderkind on Unsplash
Why is it important?
Understanding the interplay between different adaptation forms performed at different time scales (here, evolution of the body and learning of the brain) is a key step towards more autonomous robotic systems. In facts, adaptation at different time scale may enable robots to successfully face environmental changes without the intervention of human operators.
Perspectives
In this work, we applied the experience we built on voxel-based soft robots using our simulator to a different simulator. This allowed us to include reinforcement learning. Seeing how a different team designed a software for simulating voxel-based soft robots was a supplementary side value of this work.
Eric Medvet
Universita degli Studi di Trieste
Read the Original
This page is a summary of: How the Morphology Encoding Influences the Learning Ability in Body-Brain Co-Optimization, July 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3583131.3590429.
You can read the full text:
Contributors
The following have contributed to this page
