💻 computer science

Model-based Optimization of Anguilliform Swimming Gaits for Soft Robotic Applications

This paper presents the design, modeling, and optimization of the Soft Lamprey-Inspired Dual Environment Robot (SLIDER), utilizing Lighthill's theory and a genetic algorithm to co-optimize swimming gaits and caudal fin geometry, achieving a tethered speed of 0.59 body lengths per second while investigating multimodal swimming and climbing capabilities.

Original authors: Brian Van Stratum, James Gallentine, Caleb Rucker, Eric Barth, Jonathan E. Clark, Kourosh Shoele

Published 2026-06-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Brian Van Stratum, James Gallentine, Caleb Rucker, Eric Barth, Jonathan E. Clark, Kourosh Shoele

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to build a robot that can do two very different jobs: swim like a fish through water and climb up a wall like a gecko. The challenge is that what makes a robot good at swimming (being floppy and wiggly) often makes it bad at climbing (needing to be stiff and strong), and vice versa. It's like trying to design a single shoe that is perfect for running a marathon and perfect for rock climbing; the features that help in one environment might hurt in the other.

This paper introduces a new robot called SLIDER (Soft Lamprey-Inspired Dual Environment Robot). It is inspired by the Pacific lamprey, a real fish that swims in rivers and climbs up waterfalls. The team didn't just build the robot; they built a virtual "digital twin" of it to figure out the best way to make it move without having to build and break hundreds of physical prototypes.

Here is a breakdown of their work in simple terms:

1. The "Digital Twin" (The Model)

Building a robot that moves in water is hard because water pushes back in complicated ways.

The Problem: If you try to simulate the water perfectly, it takes a supercomputer days to calculate one second of movement. If you make the simulation too simple, it gives wrong answers.
The Solution: The team created a "Goldilocks" model. It's not too simple and not too complex. They used a mathematical theory (Lighthill's theory) that treats the robot like a long, flexible noodle moving through water.
How it works: The model accounts for two main forces:
- Drag (The "Mud" Effect): At slow speeds, the water feels thick and sticky, like moving through mud.
- Inertia (The "Whiplash" Effect): At fast speeds, the water feels heavy and resists being pushed aside, like the feeling of wind hitting your face when you stick your hand out of a fast car.
The Result: This model is fast enough to run on a regular computer in real-time, allowing the team to test thousands of designs instantly.

2. The "Training Wheels" (Validation)

Before trusting their digital model, they had to prove it was right.

They built the physical SLIDER robot and tied it to a sensor in a water tank so it couldn't actually swim away, but could wiggle its tail.
They compared the forces the robot felt in the real tank against the forces predicted by their computer model.
The Match: The computer and the real world agreed very closely (about 95% accuracy). This proved their "digital twin" was a reliable crystal ball for predicting how the robot would behave.

3. The "Tuning Knob" (Optimization)

Once they trusted the model, they used a Genetic Algorithm. Think of this as a digital evolution simulator.

Instead of a human guessing the best tail length or pressure, the computer "bred" thousands of virtual robot designs.
It kept the designs that swam fastest and "killed off" the slow ones, mixing their traits to create even better versions over 100 generations.
The Discovery: The computer found that to swim fast, the robot needed to wiggle its tail very quickly (high frequency) with a specific timing delay between the front and back of its body. It also found that a slightly stiffer tail worked better than a super-floppy one.

4. The Big Win (Results)

Using this digital optimization, they tuned the real robot.

The Speed: The optimized robot swam at about 21.7 cm per second. That's roughly 0.6 times its own body length every second.
The Efficiency: They found that swimming fast is different from swimming efficiently. The robot was optimized for speed, which meant it used a lot of energy, similar to how a sprinter runs fast but gets tired quickly, rather than a marathon runner who conserves energy.

5. The "Juggling Act" (Multimodal Design)

Finally, they looked at the robot's dual nature: swimming vs. climbing.

They created a "balance scale" in their computer. On one side was swimming speed; on the other was climbing ability.
They showed that if you want the robot to be a great swimmer, you need a specific tail shape. If you want it to be a great climber, you need a different shape.
The Trade-off: There is no single "perfect" robot that is the absolute best at both. You have to find a middle ground. Their model helps engineers decide exactly how much to compromise on swimming to gain climbing ability, or vice versa.

Summary

In short, the authors built a smart computer model that acts like a flight simulator for a soft robot. They used this simulator to "evolve" the best possible swimming style for their robot, SLIDER. They proved the model works by testing it in a real water tank, and they successfully used it to make the robot swim faster than it ever could have with human guesswork. They also showed how to balance the robot's design so it can handle both water and walls, even though those two tasks usually require opposite designs.

Technical Summary: Model-based Optimization of Anguilliform Swimming Gaits for Soft Robotic Applications

Problem Statement
Designing soft, bio-inspired robots capable of multimodal locomotion (specifically swimming and climbing) presents significant challenges due to the complex interactions between active actuation and passive flexible components. A primary difficulty lies in selecting a computational model that is sufficiently accurate to capture fluid-structure interactions (FSI) yet computationally efficient enough for design and control co-optimization. High-fidelity Computational Fluid Dynamics (CFD) models are often too expensive for iterative optimization, while simplified kinematic models fail to account for the feedback between fluid forces and body deformation. Furthermore, balancing conflicting physical demands—where structural features beneficial for climbing may degrade swimming performance—requires an analysis procedure capable of navigating a high-dimensional design space that is intractable for experimental direct search alone.

Methodology
The authors propose a mid-fidelity mathematical model to address these challenges, applied to the Soft Lamprey-Inspired Dual Environment Robot (SLIDER). The methodology integrates three core components:

Computational Modeling:
- Structural Physics: The robot's body is modeled using a nonlinear Euler–Bernoulli beam formulation based on Tjavaras et al., employing an Euler-Lagrangian dual coordinate system. This accounts for geometric and material nonlinearity, including a linear viscoelastic framework to capture material strain storage and pneumatic damping.
- Fluid Dynamics: The fluid environment is modeled using Lighthill's Large-Amplitude Elongated Body Theory (LAEBT). The total fluid force is decomposed into an added mass term (inertial effects) and a viscous drag term (resistive effects).
- Actuation Dynamics: To ensure model fidelity, the pneumatic pressure response is modeled as a first-order system, accounting for hydraulic delays and manifold time constants, rather than using idealized step inputs.
- Solver: The coupled fluid-structure interaction equations are solved implicitly using an efficient second-order box method, achieving a computational complexity of $O(N_p)$ per time step.
Experimental Validation:
- Experiments were conducted in a water tank using a head-fixed configuration to measure reaction forces and a tethered swimming configuration to measure velocity.
- Hydrodynamic parameters, specifically the drag coefficient ( $C_d$ ) and added mass coefficient ( $m_a$ ), were identified and validated against experimental thrust and tail excursion data. The study notes that these parameters are frequency-dependent, transitioning from drag-dominated (low frequency) to inertia-dominated (high frequency) regimes.
Co-Optimization:
- A genetic algorithm (GA) was employed to co-optimize the swimming control pattern (actuation amplitude, frequency, and phase lag) and caudal fin design (tail length and stiffness).
- A multimodal objective function was formulated to balance efficiency between swimming and climbing tasks, weighted by a parameter $\beta_{s/c}$ .

Key Contributions
The paper makes two primary contributions:

Novel Modeling Approach: The authors present a fast, accurate dynamic model coupling nonlinear large-amplitude beam theory with LAEBT. This model captures the transfer function between pressure-controlled actuation and free-swimming motion in long, slender soft robots, validated through forward dynamic FSI simulations.
Design and Control Co-Optimization: The model is applied to co-optimize the SLIDER robot's morphology and control strategy. The study demonstrates that despite the robot's morphology being optimized for climbing, the proposed modeling approach successfully identifies parameters that yield strong swimming performance, effectively managing the trade-offs inherent in multimodal design.

Results

Model Validation: The computational model showed strong agreement with experimental data. For head-fixed tests, the model achieved an $R^2$ value of 0.95 for thrust prediction across frequencies. Time-series comparisons of thrust and recoil forces at 1 Hz and 4.8 Hz confirmed the model's ability to capture both resistive and inertial flow regimes.
Swimming Performance: Using the GA, the authors co-optimized the robot to achieve a tethered swimming speed of 21.7 ± 0.4 cm/s (0.59 body lengths per second). This represents a 178% improvement in swimming performance over previous configurations.
Physical Insights:
- Low-frequency swimming is dominated by resistive environmental forces (drag), whereas higher-frequency swimming is primarily affected by inertial fluid forces.
- The optimal gait for SLIDER features a Strouhal number of approximately 2.12, indicating an optimization for thrust generation (speed) rather than cruise efficiency (typically $St \approx 0.3$ ).
- The optimal phase delay between front and rear chamber actuation was found to be 0.14, significantly different from the 0.5 (180°) often assumed in simpler models.
Multimodal Trade-offs: The study illustrates how optimal tail geometry and actuation parameters shift rapidly depending on the weighting between swimming and climbing efficiency, highlighting the necessity of a unified optimization framework.

Significance
The paper claims that its primary significance lies in providing a validated, efficient dynamic FSI model that bridges the gap between simplified kinematic approaches and expensive CFD simulations. By successfully applying this model to co-optimize a multimodal robot, the authors demonstrate that it is possible to design soft robotic platforms that perform effectively across conflicting environmental constraints (swimming vs. climbing) without relying solely on trial-and-error experimentation. The work establishes a foundation for future research into robust multimodal design criteria and the modeling of transitions between different locomotion modes.