Related Experiment Video
Updated: Sep 29, 2025
08:24
Bioinspired Soft Robot with Incorporated Microelectrodes
Published on: February 28, 2020
8.9K
Design, Modeling, and Visual Learning-Based Control of Soft Robotic Fish Driven by Super-Coiled Polymers.
Sunil Kumar Rajendran1, Feitian Zhang2
1Department of Electrical and Computer Engineering, Volgenau School of Engineering, George Mason University, Fairfax, VA, United States.
Frontiers in Robotics and AI
|March 21, 2022
Summary
This study introduces a novel soft robotic fish using super-coiled polymer artificial muscles for efficient underwater locomotion. A learning-based control system, utilizing deep-deterministic policy gradient, enables advanced path-following capabilities for aquatic exploration.
Area of Science:
- Robotics
- Bio-inspired Engineering
- Artificial Intelligence
Background:
- Aquatic bio-inspired soft robotics leverage underwater animal mechanisms for diverse applications like exploration and monitoring.
- Enhanced maneuverability and locomotion in soft robots require higher biomimicry, simplified modeling, and robust nonlinear controllers.
- Super-coiled polymers (SCP) offer advantages in flexibility, cost, and fabrication for artificial muscles, with rapid water-based cooling.
Purpose of the Study:
- To present a novel soft robotic fish design actuated by super-coiled polymers (SCP) and passively propelled by a caudal fin.
- To develop a mathematical model for the 2D swimming motion of the SCP-driven soft robotic fish.
- To design and evaluate a learning-based control strategy for yaw control and path following.
Main Methods:
- A 3-link representation was used to model the soft robotic fish's geometric and dynamic perspectives, combining SCP actuator dynamics and fish hydrodynamics.
- A deep-deterministic policy gradient (DDPG) reinforcement learning algorithm was employed for control design.
- Overhead image-based observations processed by convolutional neural networks (CNNs) were used to deduce robot curvature dynamics, bypassing complex embedded sensors. A linear quadratic regulator (LQR) provided a multi-objective reward for training.
Main Results:
- The study successfully modeled the nonlinear dynamics of the SCP-driven soft robotic fish for 2D swimming motion.
- A DDPG-based learning control design was implemented and simulated for yaw control and path following.
- The use of CNNs with overhead imagery effectively deduced the soft robot's curvature dynamics, addressing sensing limitations.
Conclusions:
- The developed soft robotic fish, actuated by SCP artificial muscles, demonstrates promising capabilities for underwater locomotion.
- The proposed learning-based control strategy, using DDPG and CNNs, effectively achieves yaw control and path following, mimicking cognitive abilities.
- This research contributes to advancing bio-inspired soft robotics with efficient actuation and intelligent control for aquatic applications.
