Automated Coadaptation of Control and Design for Friction Plates in Snake-inspired Robots
Loading...
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Author
Date
2024-08-19
Department
Major/Subject
Control, Robotics, and Autonomous Systems
Mcode
ELEC3025
Degree programme
AEE - Master’s Programme in Automation and Electrical Engineering (TS2013)
Language
en
Pages
73
Series
Abstract
Coadaptation, inspired by biological processes, allows for the simultaneous optimization of both control and design in robots. Currently, many coadaptation methods require a large population of designs, making them only testable in a simulation environment. However, with the advent of new methods in reinforcement learning, data efficient approaches to coadaptation have emerged. These methods show promise for real-world coadaptation implementation with minimal design iterations. Despite this potential, real world implementation of coadaptation remains largely unstudied. This thesis aims to study the application of coadaptation in the real world by developing the software and hardware framework necessary to complete the coadaptation process on a robotic system. A developed snake robot is the morphological focus with its interesting reliance on frictional interactions to locomote. Initial experimentation identified design parameters suitable for optimization within coadaptation. Novel bioinspired snake scales were designed for this work and demonstrated effects on snake locomotion. Reinforcement learning methods were employed to automate the continual learning of both design and control. Through a hyperparameter ablation study, a robust deep reinforcement learning algorithm was formulated, capable of learning locomotion across different designs. The aforementioned results were integrated into a real world coadaptation framework. This framework demonstrated the feasibility of executing coadaptation in the real world on frictional scales of a snake robot. While more design iterations with this framework may still be required to observe improved locomotion, the developed framework serves as the foundation for future snake robot coadaptation experimentation and provides a base model for other real world coadaptation implementations.Description
Supervisor
Kyrki, VilleThesis advisor
Chaubey, ShivamLuck, Kevin Sebastian
Keywords
robot design, deep reinforcement learning, snake robot, coadaptation, real-world robot implementation
