Abstract
Soft robotics has the potential to revolutionize robotic locomotion, in particular, soft robotic swimmers offer a minimally invasive and adaptive solution to explore and preserve our oceans. Unfortunately, current soft robotic swimmers are vastly inferior to evolved biological swimmers, especially in terms of controllability, efficiency, maneuverability, and longevity. Additionally, the tedious iterative fabrication and empirical testing required to design soft robots has hindered their optimization. In this work, we tackle this challenge by providing an efficient and straightforward pipeline for designing and fabricating soft robotic swimmers equipped with electrostatic actuation. We streamline the process to allow for rapid additive manufacturing, and show how a differentiable simulation can be used to match a simplified model to the real deformation of a robotic swimmer. We perform several experiments with the fabricated swimmer by varying the voltage and actuation frequency of the swimmer's antagonistic muscles. We show how the voltage and frequency vary the locomotion speed of the swimmer while moving in liquid oil and observe a clear optimum in forward swimming speed. The differentiable simulation model we propose has various downstream applications, such as control and shape optimization of the swimmer; optimization results can be directly mapped back to the real robot through our sim-to-real matching.
Introduction
Soft robots are uniquely suited for marine exploration due to their compliance, resilience, and ability to mimic the efficient locomotion of biological organisms. However, designing these systems is notoriously difficult. Unlike rigid-body robots, soft robots undergo complex, non-linear deformations that are challenging to model accurately. Traditionally, engineers have relied on a "trial and error" approach—manually building, testing, and tweaking designs—which is both time-consuming and expensive.
This research addresses the "sim-to-real gap," where digital models often fail to capture the nuances of physical materials and actuators. By combining automated multimaterial 3D printing with a differentiable simulation framework, the authors present a pipeline that allows a digital twin to be automatically tuned to match the behavior of a physical soft robotic swimmer.
Design and Fabrication
The swimmer's body is produced using a tool-changing FDM (Fused Deposition Modeling) printer, allowing for a single-step manufacturing process of complex, bio-inspired geometries. The design utilizes two distinct materials: rigid polypropylene (PP) for the "spine" and soft styrene-ethylene-butylene-styrene (SEBS) for the flexible body components.
Actuation is provided by Peano-HASEL (Hydraulically Amplified Self-healing ELectrostatic) actuators. These "electrostatic muscles" consist of oil-filled pouches that contract when high voltage is applied. They are lightweight, fast, and mimic the contractile properties of biological muscle fibers, making them ideal for high-performance soft robotics.
Differentiable Simulation Framework
The core innovation of this work is the use of a differentiable Finite Element Method (FEM) simulator, known as DiffPD. Unlike standard simulators, a differentiable one allows the system to calculate gradients. This means the optimization algorithm can "see" how changing a specific parameter—such as material stiffness or muscle strength—directly affects the resulting swimming performance.
To simplify the complex physics of the HASEL actuators, the researchers implemented a simplified "Muscle Model" within the simulation. By treating the simulation parameters as variables in an optimization problem, they used the Adam optimizer to minimize the Mean Squared Error (MSE) between the simulated bending angles and the real-world data.
Experimental Results and Sim-to-Real Matching
The researchers conducted experiments with the swimmer in a canola oil medium. Canola oil serves two purposes: it acts as an electrical insulator for the high-voltage actuators and provides a low-Reynolds-number environment that simplifies fluid dynamics modeling.
System Identification
To calibrate the model, the team generated a targeted dataset:
- Training: Video recordings of the swimmer at varying voltages (3.5, 4.5, 5.5 kV) and frequencies (0.5, 2.0, 3.5 Hz).
- Tracking: The CSRT (Channel and Spatial Reliability Tracking) algorithm was used to extract precise bending angles from the footage.
- Validation: The model was tested against unseen cases (2.5 Hz at 4.25, 4.8, and 5.2 kV) to ensure generalizability.
Performance Metrics
The optimized simulation achieved remarkable accuracy, with a mean absolute angle error of $1.553 \times 10^{-2}$ radians for training data and $3.103 \times 10^{-2}$ radians for test data. Furthermore, the experiments identified a clear performance optimum: the swimmer reached its maximum forward speed at an actuation frequency of 7 Hz across all tested voltages.
Limitations and Future Outlook
While the current pipeline represents a significant step forward, several challenges remain. The swimmer is currently tethered to a high-voltage power supply, and the simulation does not yet perfectly capture the material damping (energy loss) observed at high frequencies.
Future work will focus on developing an untethered, autonomous version of the swimmer with onboard electronics. Additionally, the researchers aim to integrate a learned hydrodynamics model to simulate the complex interactions between the soft body and the surrounding fluid, enabling full end-to-end shape and control optimization.
Summary updated