AINEWS
One Soft Body, Two Domains: Co-Designing a Lamprey Robot to Swim and Climb
A validated Lighthill based fluid model and a genetic algorithm jointly shaped SLIDER's caudal fin and gait, treating water and walls as one design problem.
A validated Lighthill based fluid model and a genetic algorithm jointly shaped SLIDER's caudal fin and gait, treating water and walls as one design problem.
Most soft robots are designed for one environment at a time. A new preprint argues the better question is how a single soft body should be built when it has to perform in two, and answers it with a lamprey-inspired swimmer-climber called SLIDER.
The Model-based Optimization of Anguilliform Swimming Gaits for Soft Robotic Applications preprint introduces SLIDER, the Soft Lamprey-Inspired Dual Environment Robot, and treats swimming and climbing as one co-design problem. Instead of tuning a gait for water and a body for walls separately, the authors co-optimize the robot's caudal fin and its swimming control pattern under the morphology constraints shared with a climbing gait on the same body.
The substrate for that search is a fluid-structure interaction model rather than a learned policy. The fluid side uses Lighthill's large-amplitude elongated body theory, which captures inertial, vortex, and viscous effects for an undulating body. The structural side is geometrically and materially nonlinear, accounting for internal pressure, tail size, and body stiffness, and is validated against the physical robot. The coupled problem is solved implicitly with a second-order box method for efficiency, then driven by a genetic algorithm that searches the co-design space within SLIDER's physical and actuation limits, as described in the arXiv paper.
The result is a tethered swimming speed of 21.7 ± 0.4 cm/s, or 0.59 body lengths per second, for a body that is also being studied as a wall climber. The same optimization procedure is extended to a multimodal swimming-and-climbing study, framing the work as design methodology for soft robots that have to cross physical regimes rather than a swimming benchmark.
The regime split is itself a contribution. At low swimming frequencies, the environment is dominated by resistive (viscous and pressure) forces. At higher frequencies, inertial fluid forces take over. Crossing that boundary is the kind of design choice a co-optimization framework is meant to surface, because the same morphology cannot be tuned for both ends of the spectrum at once. Designing for the climb, in turn, constrains the swimmer.
Two limits are worth naming. SLIDER swims tethered in a quiescent water tank, so the reported speed reflects an idealized setup rather than field conditions. Lighthill's elongated-body theory carries its own assumptions at large amplitude, and the climbing gait is investigated rather than fully co-optimized end-to-end with the swimming gait. The work is also a single arXiv preprint, with findings self-reported by the authors and the FSI validation and genetic-algorithm gains not yet independently reproduced.
What to watch: whether the Lighthill-plus-implicit-solver stack holds up outside a quiescent tank, and whether the climbing gait can be folded into the co-optimization loop so that one body truly negotiates both domains rather than just tolerating them.