Co-Design and Control of a Biomimetic Snake Robot and its Contact Surfaces with Reinforcement Learning

Liza Darwesh, Riccardo Pretto, Sylvia Cressman, Tiia Tikkala, Shivam Chaubey, Kevin Sebastian Luck

Keywords: Robotics, Reinforcement Learning, Co-Design, Bioinsipred, Biomimetic

Abstract:

Snake robots offer promising capabilities for locomotion in complex environments and terrain due to their modular structure and distributed actuation.
However, achieving efficient movement in the real world requires not only effective control policies but also morphological designs adapted to the terrain.
This paper investigates real-world coadaptation in snake robots by coupling Soft Actor-Critic (SAC) reinforcement learning.
Specifically, we investigate the effect on locomotion nd control of bio-inspired scale-designs on the contact-surfaces of a snake robot.
We evaluate four morphological designs across three physical terrains in the real world.
Results demonstrate that morphology significantly influences learning speed and final performance, and that certain designs generalize better across environments.
This is a first step towards bio-inspired and -mimetic snake designs utilizig optimized scales combining reinforcement learning with parameterized scale designs.

Confirmation: I understand that authors of each paper submitted to EWRL may be asked to review 2-3 other submissions to EWRL.

Serve As Reviewer: ~Liza_Darwesh1

Track: Regular Track: unpublished work

Submission Number: 173