Cross-sectional Topology Optimization of Slender Soft Pneumatic Actuators using Genetic Algorithms and Geometrically Exact Beam Models

Kristin Miriam de Payrebrune; Leon Schindler

REVIEW 2 major objections 6 minor 31 references

Cross-sectional Topology Optimization of Slender Soft Pneumatic Actuators using Genetic Algorithms and Geometrically Exact Beam Models

T0 review · 2 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A genetic algorithm over Voronoi-tessellated cross-sections finds soft pneumatic actuator designs whose simulated end-effector workspaces match target workspaces in three case studies.

desk verdict A solid simulation-level demonstration of GA-based cross-section design for soft actuators, but the scalarized fitness with λ=1000 hides whether the primary workspace objective actually drives the optimization. read the letter →

arxiv 2412.16138 v1 pith:CT5R5ZZS submitted 2024-12-20 cs.RO physics.comp-ph

Leon Schindler , Kristin Miriam de Payrebrune This is my paper · ORCID

classification cs.ROphysics.comp-ph

keywords softroboticstopologyoptimizationgeneticalgorithmCosseratbeammodelpneumaticactuatorVoronoitessellationworkspacedesignblack-box

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether a design tool can replace manual trial-and-error in soft robotics by letting an algorithm sculpt the cross-section of a slender pneumatic actuator so that its tip sweeps out a desired workspace. The authors encode each candidate design as a Voronoi tessellation of the circular cross-section, simulate the deformed backbone with a geometrically exact Cosserat beam model under chamber pressures, and evolve the tessellation with a genetic algorithm to minimize the squared distance between computed and target tip positions. Across three case studies—a random target geometry, a hand-designed geometry, and a hand-specified workspace with no underlying geometry—the optimizer produced cross-sections whose simulated workspaces track the targets, with best losses of 19.49, 15.07, and 1119.73 mm². The claim is that a black-box, simulation-guided search can suggest plausible prototypes that an operator can then refine and manufacture. The demonstration is deliberately confined to a simplified model; no prototypes were built.

What carries the argument

The central machinery is the Voronoi-tessellated cross-section used as the genotype, coupled with the geometrically exact Cosserat beam model as the phenotype simulator. A design is generated by randomly placing 100 feature points in a polar-discretized ring with 128 radial and 360 circumferential elements, assigning each point a type (material or one of three pressure chambers), and coloring every element by its nearest feature point, with a material wall enforced between different chambers. The beam model reduces the three-dimensional deformation to a one-dimensional centerline problem; chamber pressure loads are computed from each chamber's area and centroid, and the equilibrium backbone is found in the principal-axis frame while neglecting torsion, ballooning, necking, and gravity. The fitness landscape is defined by the squared-distance loss over the 27 workspace nodes plus a Delaunay-edge-based chamber-fragmentation penalty.

What would settle it

Manufacture the best cross-section from each case study with radial reinforcement to suppress ballooning, pressurize the three chamber groups at the 27 pressure combinations, measure the tip positions, and compute the same squared-distance loss; close agreement would confirm the central claim, while systematic deviation due to torsion, necking, or stiffness modeling would show the claim currently holds only inside the simulator.

Watch

Extended reading notes

Core claim

The paper's central claim is that black-box topology optimization with genetic algorithms is capable of finding cross-sectional designs for slender soft pneumatic actuators that reach a user-specified target workspace. A workspace is represented by 27 end-effector positions, computed at all combinations of pressures 0, 50, and 100 kPa for three independently supplied chamber groups. The genotype is a set of 100 Voronoi feature points, each labeled as material or as one of three chambers; the phenotype is the workspace predicted by a Cosserat beam model with pressure loads computed from the chamber areas and centroids. The objective combines the summed squared distances to the target positions with a penalty on fragmented chamber clusters, and elitist selection drives the search. With the best-performing recombination and mutation variants, the method matches geometry-derived target workspaces closely and approximates a hand-specified workspace in position and alignment, although the absolute loss for the hand-specified target remains large.

Load-bearing premise

The load-bearing premise is that the simplified beam simulation faithfully represents a physical actuator, so a design that scores well in simulation would also score well in reality; no prototype was built or measured to test this.

Editorial extensions

If this is right

If the central claim holds, a designer can specify a target workspace as a handful of pressure–position pairs and receive a concrete cross-sectional geometry without manual prototyping.
Because the beam model cuts the simulation cost from a three-dimensional volumetric problem to a one-dimensional centerline problem, many candidate designs can be screened in simulation before any physical prototype is made.
Across the two geometry-derived targets, the best losses were 19.49 and 15.07 mm², achieved by the range-weighted and fixed-direct variants respectively, while the hand-specified workspace reached a best loss of 1119.73 mm² with range-based direct mutation.
The optimized designs are virtual; any real deployment requires manufacturing, radial reinforcement against ballooning, and experimental verification, which the paper identifies as the next step.
The workspace representation via quadratic hexahedra with 27 nodes means the operator can provide a rough region rather than exact continuous shapes, and the optimizer will try to hit that region's sampled points.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial extension: because the fitness is a sum of squared tip-position deviations at 27 pressure combinations, the same pipeline should also be able to optimize toward a set of target quasi-static trajectories rather than a full workspace, a direction the authors mention.
Editorial extension: the Voronoi encoding is compact and smooth, so a natural next test is to evolve a stack of cross-sections to allow the cross-section to vary along the length; that extension would bring torsion and ballooning back into the model and could be checked against the finite-element comparison the authors propose.
Editorial extension: adding a manufacturability penalty, such as penalizing disconnected material islands or enforcing a minimum wall thickness, would reduce the operator preprocessing step the paper says remains necessary, at the cost of one more scalar weight.
Editorial extension: the large best loss for the hand-specified workspace suggests the optimizer is limited by what a constant-cross-section, torsion-free beam can physically express; a useful test is to run the same target through a model that allows torsion and see whether the loss drops.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper is a solid engineering demonstration: a GA over Voronoi-encoded cross-sections, evaluated with a Cosserat beam model, produces simulated workspaces that match target workspaces for a slender pneumatic actuator. That combination is new, and the authors are unusually forthright about the limitations—no experimental validation, simplified beam physics, no torsion or ballooning.

What the paper does well: the design representation is sensible, the beam model choice is justified, and the comparison of four recombination/mutation schemes is systematic. The results for the two geometry-derived targets are visually convincing, and the best losses are low. The authors also correctly frame case 3 as an approximate fit.

The soft spot is methodological. The fitness is T = L + K with K = 1000r, where r is the ratio of different-type to same-type Delaunay edges. For the nominal type probabilities, a random chromosome has r ≈ 2, so K ≈ 2000 mm², while the reported L values range from 15 to 150 mm² in the first two cases. That means selection is initially driven almost entirely by K, not L. The paper reports only L as the metric and concludes the GA optimizes workspace matching, but without reporting K at the optimum, it is unclear whether the low L values are a direct consequence of the objective or an incidental byproduct of a selection pressure that favors clustered designs. This is not fatal—the best designs do match the targets—but it weakens the central claim. The fix is easy: use an adaptive weighting or normalize both objectives, and report the Pareto front or at least track K alongside L.

The other limitations are stated honestly: targets for cases 1 and 2 come from the same simulator, so this is a self-consistency test rather than external validation, and there is no experimental check. For a methods paper that is acceptable, but it should temper the language that the GA "proves to be capable" of finding good designs.

Who is this for? Researchers working on automated design of soft actuators, especially those using beam models. It is a useful reference for design representations and GA settings. I would cite it in work on soft actuator co-design. It deserves a serious referee; the scalarization issue should be addressed in revision.

Referee Report

2 major / 6 minor

Summary. The manuscript presents a black-box topology-optimization method for designing the cross-sections of slender soft pneumatic actuators. A polar grid is labeled as material or one of three pressure chambers via Voronoi tessellation; the resulting actuator is simulated with a quasi-static geometrically exact Cosserat beam model with pressure loads (Sections 2.1.4-2.1.5). A genetic algorithm minimizes a scalarized fitness T = L + K (Eq. 14), where L is the squared distance of 27 end-effector positions to a target workspace (Eq. 15) and K = 1000 r penalizes the ratio of Delaunay edges connecting different feature types (Eq. 16). Three case studies target workspaces derived from a random cross-section, a hand-designed cross-section, and a hand-specified workspace. Best losses are 19.49, 15.07, and 1119.73 mm^2, and the authors conclude that the GA finds good designs for plausible targets. The paper explicitly states that the model is simplified and that no experimental validation was performed.

Significance. This is a plausible proof-of-concept for combining genetic algorithms, Voronoi-based cross-section encoding, and a geometrically exact beam model for soft-actuator design. The authors are transparent about the limitations of the simulator (Sections 2.1.4, 5), and the formulation of the workspace as 27 pressure combinations is clear. The central claim is currently supported only within a self-consistent simulation environment: the targets in Cases 1 and 2 are generated by the same forward model used for fitness evaluation. If the objective-scaling issue is fixed, the method is a useful design-assistant tool; the work would benefit from a comparison with FEM or an experiment, but the lack of such validation is not by itself disqualifying for a computational proof-of-concept.

major comments (2)

[Section 2.2.2, Eq. (16)] The secondary objective can dominate the primary objective, so the reported L values do not by themselves support the claim that the GA optimizes workspace matching. For the nominal feature-type probabilities (pm = 1/2, pc = 1/6), a random genotype has approximately P(same-type Delaunay edge) = 1/3 and P(different-type edge) = 2/3, giving r approximately 2 and K approximately 2000 mm^2. The best final losses in Table 1 are 19.49 mm^2 (Case 1), 15.07 mm^2 (Case 2), and 1119.73 mm^2 (Case 3); hence K is roughly two orders of magnitude larger than L in the first two cases and comparable in the third. Since selection ranks by T = L + K, the algorithm is predominantly minimizing the cluster-boundary ratio unless r is driven to very small values in the final populations. The manuscript reports no r or K values for the optimized individuals and no sensitivity analysis for lambda; without these, the low L values could be incidental byproducts of low-K selection. Please report the components of T during optimization, and include an ablation (e.g., lambda = 0 or a range of lambda values) to demonstrate that the primary workspace objective drives the design.
[Abstract and Sections 3 and 5] The central claim that the GA 'proves to be capable of finding good designs' is stronger than the evidence. In Case studies 1 and 2 the target workspaces are generated with the same beam model and pressure-load implementation used as the fitness function (Section 2.1.4 and Section 3), so these experiments establish internal consistency of the optimization loop rather than predictive validity for a physical actuator. Given the acknowledged simplifications (constant cross-section, no ballooning, IT = 0, no gravity, no end caps) and the absence of experimental or FEM validation, the conclusion should be phrased within the simplified model, or a higher-fidelity validation should be added.

minor comments (6)

[Eq. (15)] The sum runs from j = 0 to 27, which gives 28 terms, while the text states that the workspace is represented by 27 positions; the index range should be corrected.
[Algorithm 1] After sampling indices nr and nphi, the conversion to the feature-point coordinates (r, phi) used in the Cartesian distance computation is not specified; this is needed for reproducibility.
[Sections 2.1.2 and 2.2.3] The symbol pm is used both for the material-site probability and for the mutation ratio; these two quantities should be given distinct names.
[Section 3, Figure 7] The text says the plotted quantity is min_p L(p), but the caption says 'total loss' (T = L + K); please make the metric in the figure explicit and consistent with Table 1.
[Section 4] The statement that the method with the shortest runtime should be selected is not supported by any reported runtime measurements; a table of computational cost would support this practical recommendation.
[Section 3] The comparison of the four recombination/mutation variants is based on only four iterations per variant, and no statistical significance tests are given; given the observed spread (e.g., Case 1 range 19.49-152.52), the statement that 'all four methods deliver similar results' should be softened or supported by more runs.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the GA optimization is self-contained, with targets generated independently of the optimizer; remaining concerns are validation and objective weighting, not circular derivation.

full rationale

The paper's chain of reasoning is an optimization procedure rather than a derivation in which an output is secretly an input. The forward model is a fixed Cosserat beam formulation taken from the external reference Till et al. [27], with material constants E = 300 kPa and G = 100 kPa taken from the silicone literature, and no parameters are fitted to the target workspaces. The fitness function L in Eq. (15) compares simulator outputs h(H, P1,j, P2,j, P3,j) with target positions wj, and in Case 1 and Case 2 those targets are themselves generated by the same simulator from a known cross-section. That makes those two cases self-consistency benchmarks rather than external validations, but it is not a circularity: the target positions are not used to define the beam model, the Voronoi design encoding, or the correctness of the simulator, and the known generating cross-section is not fed into the optimizer. Case 3 uses a hand-specified target workspace, and the paper explicitly discloses that the actuators are virtual, that the model is simplified, and that no experimental validation has yet been performed. The one notable internal-validity concern is the scalarization T = L + K with K = lambda*r and lambda = 1000 (Eqs. 14-16), since the secondary cluster-boundary loss may dominate the reported primary workspace loss; however, that is an objective-weighting and correctness concern, not a circular-reduction of a claimed result into its own inputs. No load-bearing self-citation, imported uniqueness theorem, ansatz smuggled in via citation, or renaming of a known result is present.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The central result rests on the beam model being an adequate stand-in for a physical actuator, on the chosen Voronoi and GA encoding parameters, and on the 27-point workspace description. No new physical entities are introduced. The free parameters are algorithm settings chosen by hand rather than fitted to external data.

free parameters (6)

Secondary objective weight lambda = 1000
Eq. 16 sets K = lambda * r with lambda = 1000, chosen with no sensitivity analysis; since r is a ratio of mixed-type to same-type Delaunay edges, lambda can make K dominate the primary loss L in Eq. 14.
Voronoi feature point count Nf = 100
Number of feature points in the cross-section encoding; chosen ad hoc, not varied.
Feature type probabilities pm, pc = pm = 1/2, pc = 1/6
Assigned to material and three chamber types; chosen by hand to balance stiffness and chamber area.
Wall thickness = 4 elements
Post-generation material wall between different chambers; not constant in physical thickness because radial discretization is non-uniform.
GA hyperparameters = Np=20, Ng=100, pc=0.8, nu=0.5, pm=0.1, pf=0.1
Population, generations, crossover and mutation rates chosen without sensitivity analysis; the central claim depends on these settings.
Discretization resolution = Nr=128, Nphi=360
Cross-section grid resolution chosen for computational cost; no convergence study is reported.

assumptions (7)

domain assumption Cosserat beam model with linear-elastic isotropic material and pressure loads from Till et al. accurately describes slender soft pneumatic actuator deformation.
Used in Sections 2.1.4 and 2.1.5; the optimizer's fitness is computed from this model, so its accuracy is load-bearing.
domain assumption Incompressible material with no necking and constant cross-sectional area; no radial reinforcement needed.
Stated in Section 2.1.4 and the conclusion; ballooning would change pressurized area and pressure loads.
domain assumption Torsional stiffness neglected (IT = 0) and shear correction kappa = 1.
Section 2.1.4; justified only for constant cross-section and no external torsional torques.
domain assumption Quasi-static pressurization, no gravity, and no external loads.
Section 2.1.4; dynamic terms are set to zero and fe = le = 0.
domain assumption 27 pressure combinations and quadratic hexahedra approximate the continuous workspace.
Section 2.1.7; the objective function in Eq. 15 only measures these 27 points, so the true workspace may deviate between samples.
ad hoc to paper For case studies 1 and 2, workspaces generated by the same forward beam model are valid ground-truth targets.
Evaluation is self-consistency within the simulator, not external validation; case 3 uses a hand-specified target.
standard math Voronoi tessellation and Delaunay triangulation algorithms correctly implement the geometric operations described.
Sections 2.1.2 and 2.2.2 rely on them; these are standard computational geometry algorithms with low risk.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Cross-sectional Topology Optimization of Slender Soft Pneumatic Actuators using Genetic Algorithms and Geometrically Exact Beam Models." pith.science (2026). https://pith.science/paper/CT5R5ZZS

@misc{pith2026241216138,
  author       = {Pith},
  title        = {Pith review of: Cross-sectional Topology Optimization of Slender Soft Pneumatic Actuators using Genetic Algorithms and Geometrically Exact Beam Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CT5R5ZZS}},
  note         = {Machine review of arXiv:2412.16138}
}

read the original abstract

The design of soft robots is still commonly driven by manual trial-and-error approaches, requiring the manufacturing of multiple physical prototypes, which in the end, is time-consuming and requires significant expertise. To reduce the number of manual interventions in this process, topology optimization can be used to assist the design process. The design is then guided by simulations and numerous prototypes can be tested in simulation rather than being evaluated through laborious experiments. To implement this simulation-driven design process, the possible design space of a slender soft pneumatic actuator is generalized to the design of the circular cross-section. We perform a black-box topology optimization using genetic algorithms to obtain a cross-sectional design of a soft pneumatic actuator that is capable of reaching a target workspace defined by the end-effector positions at different pressure values. This design method is evaluated for three different case studies and target workspaces, which were either randomly generated or specified by the operator of the design assistant. The black-box topology optimization based on genetic algorithms proves to be capable of finding good designs under given plausible target workspaces. We considered a simplified simulation model to verify the efficacy of the employed method. An experimental validation has not yet been performed. It can be concluded that the employed black-box topology optimization can assist in the design process for slender soft pneumatic actuators. It supports at searching for possible design prototypes that reach points specified by corresponding actuation pressures. This helps reduce the trial-and-error driven iterative manual design process and enables the operator to focus on prototypes that already offer a good viable solution.

Figures

Figures reproduced from arXiv: 2412.16138 by the authors.

**Figure 2.** Discretized circular cross-section of a soft pneumatic actuator generated by a random Voronoi tessellation [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** a) Randomized circular cross-section with the corresponding workspace of the end-effector rendering in top-down view, and b) the same cross-section and workspace rendered in a diagonal view. 2.1.5 Pressure loads In extension to the Cosserat beam model described in the previous section, pressure loads are applied to the model of the soft pneumatic actuator. These pressure loads are derived in Till et al. [27] and des… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: First case study with a random cross-section to generate target workspace. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 5.** Figure 5: Second case study with a specific cross-section to generate target workspace. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

**Figure 6.** Figure 6: Third case study with a manually designed target workspace without being computed from a simulation using [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

**Figure 7.** Figure 7: Minimum, mean, and maximum total loss of the best individual in each generation over all iterations for each [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 19 canonical work pages

[1]

Design, Fabrication and Control of Soft Robots

Daniela Rus and Michael T. Tolley. “Design, Fabrication and Control of Soft Robots”. In: Nature 521.7553 (May 28, 2015), pp. 467–475. ISSN : 0028-0836, 1476-4687. DOI: 10.1038/nature14543

doi:10.1038/nature14543 2015
[2]

Challenges and Opportunities for Design, Simulation, and Fabrication of Soft Robots

Hod Lipson. “Challenges and Opportunities for Design, Simulation, and Fabrication of Soft Robots”. In: Soft Robotics 1.1 (Mar. 2014), pp. 21–27. ISSN : 2169-5172, 2169-5180. DOI: 10.1089/soro.2013.0007

arXiv 2014
[3]

Soft Actuators for Soft Robotic Applications: A Review

Nazek El-Atab, Rishabh B. Mishra, Fhad Al-Modaf, Lana Joharji, Aljohara A. Alsharif, Haneen Alamoudi, Marlon Diaz, Nadeem Qaiser, and Muhammad Mustafa Hussain. “Soft Actuators for Soft Robotic Applications: A Review”. In: Advanced Intelligent Systems 2.10 (Oct. 2020), p. 2000128. ISSN : 2640-4567, 2640-4567. DOI: 10.1002/aisy.202000128

doi:10.1002/aisy.202000128 2020
[4]

From Bioinspiration to Computer Generation: Developments in Autonomous Soft Robot Design

Joshua Pinskier and David Howard. “From Bioinspiration to Computer Generation: Developments in Autonomous Soft Robot Design”. In: Advanced Intelligent Systems 4.1 (Jan. 2022), p. 2100086. ISSN : 2640-4567, 2640-4567. DOI: 10.1002/aisy.202100086

work page doi:10.1002/aisy.202100086 2022
[5]

Evolutionary Black-Box Topology Optimization: Challenges and Promises

David Guirguis, Nikola Aulig, Renato Picelli, Bo Zhu, Yuqing Zhou, William Vicente, Francesco Iorio, Markus Olhofer, Wojciech Matusik, Carlos Artemio Coello Coello, and Kazuhiro Saitou. “Evolutionary Black-Box Topology Optimization: Challenges and Promises”. In: IEEE Transactions on Evolutionary Computation 24.4 (Aug. 2020), pp. 613–633. ISSN : 1089-778X,...

work page arXiv 2020
[6]

Design and Control of Soft Robots Using Dif- ferentiable Simulation

Moritz Bächer, Espen Knoop, and Christian Schumacher. “Design and Control of Soft Robots Using Dif- ferentiable Simulation”. In: Current Robotics Reports 2.2 (June 2021), pp. 211–221. ISSN : 2662-4087. DOI: 10.1007/s43154-021-00052-7

work page doi:10.1007/s43154-021-00052-7 2021
[7]

Towards Automatic Design of Soft Pneumatic Actuators: Inner Structure Design Using CNN Model and Bézier Curve-Based Genetic Algorithm

Loïc Mosser, Laurent Barbé, Lennart Rubbert, and Pierre Renaud. “Towards Automatic Design of Soft Pneumatic Actuators: Inner Structure Design Using CNN Model and Bézier Curve-Based Genetic Algorithm”. In: IEEE Robotics and Automation Letters 8.10 (Oct. 2023), pp. 6603–6610. ISSN : 2377-3766, 2377-3774. DOI: 10.1109/ LRA.2023.3309135

arXiv 2023
[8]

Design and Development of a Soft Gripper with Topology Optimization

Hongying Zhang, Michael Yu Wang, Feifei Chen, Yiqiang Wang, A. Senthil Kumar, and Jerry Y . H. Fuh. “Design and Development of a Soft Gripper with Topology Optimization”. In: 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Vancouver, BC: IEEE,...

work page arXiv 2017

Show all 31 references

[9]

Topology Optimized Multimaterial Soft Fingers for Applications on Grippers, Rehabilitation, and Artificial Hands

Hongying Zhang, A. Senthil Kumar, Feifei Chen, Jerry Y . H. Fuh, and Michael Yu Wang. “Topology Optimized Multimaterial Soft Fingers for Applications on Grippers, Rehabilitation, and Artificial Hands”. In: IEEE/ASME Transactions on Mechatronics 24.1 (Feb. 2019), pp. 120–131. I...

2019
[10]

A Computational Design Framework for Pressure-driven Soft Robots through Nonlinear Topology Optimization

Brandon Caasenbrood, Alexander Pogromsky, and Henk Nijmeijer. “A Computational Design Framework for Pressure-driven Soft Robots through Nonlinear Topology Optimization”. In:2020 3rd IEEE International Con- ference on Soft Robotics (RoboSoft). 2020 3rd IEEE International Confer...

2020
[11]

Optimal Design of Soft Pneumatic Bending Actuators Subjected to Design-Dependent Pressure Loads

Yongyu Chen, Zeyang Xia, and Qunfei Zhao. “Optimal Design of Soft Pneumatic Bending Actuators Subjected to Design-Dependent Pressure Loads”. In: IEEE/ASME Transactions on Mechatronics 24.6 (Dec. 2019), pp. 2873–

2019
[12]

Design Optimization of Soft Pneumatic Actuators Using Genetic Algorithms

Gundula Runge, Jan Peters, and Annika Raatz. “Design Optimization of Soft Pneumatic Actuators Using Genetic Algorithms”. In: 2017 IEEE International Conference on Robotics and Biomimetics (ROBIO) . 2017 IEEE International Conference on Robotics and Biomimetics (ROBIO). Macau: ...

2017
[13]

Evolving Amorphous Robots

Jonathan D. Hiller and Hod Lipson. “Evolving Amorphous Robots”. In: IEEE Symposium on Artificial Life . 2010

2010
[14]

Morphological Evolution of Freeform Robots

Jonathan D. Hiller and Hod Lipson. “Morphological Evolution of Freeform Robots”. In: Proceedings of the 12th Annual Conference on Genetic and Evolutionary Computation. GECCO ’10: Genetic and Evolutionary Computation Conference. Portland Oregon USA: ACM, July 7, 2010, pp. 151–1...

2010
[15]

Multi Material Topological Optimization of Structures and Mechanisms

Jonathan D. Hiller and Hod Lipson. “Multi Material Topological Optimization of Structures and Mechanisms”. In: Proceedings of the 11th Annual Conference on Genetic and Evolutionary Computation. GECCO09: Genetic and Evolutionary Computation Conference. Montreal Québec Canada: A...

2009
[16]

Automatic Design and Manufacture of Soft Robots

Jonathan Hiller and Hod Lipson. “Automatic Design and Manufacture of Soft Robots”. In: IEEE Transactions on Robotics 28.2 (Apr. 2012), pp. 457–466. ISSN : 1552-3098, 1941-0468. DOI: 10.1109/TRO.2011.2172702

2012
[17]

Differentiable Soft-Robot Generation

François Cochevelou, David Bonner, and Martin-Pierre Schmidt. “Differentiable Soft-Robot Generation”. In: Proceedings of the Genetic and Evolutionary Computation Conference. GECCO ’23: Genetic and Evolutionary Computation Conference. Lisbon Portugal: ACM, July 15, 2023, pp. 12...

2023
[18]

Efficient Automatic Design of Robots

David Matthews, Andrew Spielberg, Daniela Rus, Sam Kriegman, and Josh Bongard. “Efficient Automatic Design of Robots”. In: Proceedings of the National Academy of Sciences 120.41 (Oct. 10, 2023), e2305180120. ISSN : 0027-8424, 1091-6490. DOI: 10.1073/pnas.2305180120

2023 doi
[19]

Soft Pneumatic Actuator Design Using Differentiable Simulation

Arvi Gjoka, Espen Knoop, Moritz Bächer, Denis Zorin, and Daniele Panozzo. “Soft Pneumatic Actuator Design Using Differentiable Simulation”. In: Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers ’24. SIGGRAPH ’24: Special Intere...

2024
[20]

Best Practices for Differentiable Soft Robot Modeling and Optimization with the Material Point Method

Kam Bielawski, Jakub Rozlivek, Matej Hoffmann, and Josh Bongard. “Best Practices for Differentiable Soft Robot Modeling and Optimization with the Material Point Method”. In: ALIFE 2024: Proceedings of the 2024 Artificial Life Conference. July 1, 2024, p. 11. DOI: 10.1162/isal_a_00724

2024 doi
[21]

Creation of Novel Soft Robot Designs Using Generative AI

Wee Kiat Chan, PengWei Wang, and Raye Chen-Hua Yeow. Creation of Novel Soft Robot Designs Using Generative AI. May 3, 2024. DOI: 10.48550/arXiv.2405.01824 . arXiv: 2405.01824 [cs] . URL: http: //arxiv.org/abs/2405.01824 (visited on 10/28/2024). Pre-published

2024 doi arXiv
[22]

Dimension Optimization of Pneumatically Actuated Soft Continuum Manipulators

Xiangyu Peng, Ningbin Zhang, Lisen Ge, and Guoying Gu. “Dimension Optimization of Pneumatically Actuated Soft Continuum Manipulators”. In: 2019 2nd IEEE International Conference on Soft Robotics (RoboSoft) . 2019 2nd IEEE International Conference on Soft Robotics (RoboSoft). S...

2019
[23]

Computational Design and Fabrication of Soft Pneumatic Objects with Desired Deformations

Li-Ke Ma, Yizhonc Zhang, Yang Liu, Kun Zhou, and Xin Tong. “Computational Design and Fabrication of Soft Pneumatic Objects with Desired Deformations”. In: ACM Transactions on Graphics 36.6 (Dec. 31, 2017), pp. 1–12. ISSN : 0730-0301, 1557-7368. DOI: 10.1145/3130800.3130850

2017
[24]

Design of Soft Multi-Material Pneumatic Actuators Based on Principal Strain Field

Longwei Ding, Ning Dai, Xiaoming Mu, Shaohui Xie, Xu Fan, Dawei Li, and Xiaosheng Cheng. “Design of Soft Multi-Material Pneumatic Actuators Based on Principal Strain Field”. In: Materials & Design 182 (Nov. 2019), p. 108000. ISSN : 02641275. DOI: 10.1016/j.matdes.2019.108000

2019
[25]

Automated Routing of Muscle Fibers for Soft Robots

Guirec Maloisel, Espen Knoop, Christian Schumacher, and Moritz Bacher. “Automated Routing of Muscle Fibers for Soft Robots”. In: IEEE Transactions on Robotics 37.3 (June 2021), pp. 996–1008. ISSN : 1552-3098, 1941-0468. DOI: 10.1109/TRO.2020.3043654

2021
[26]

V oronoi Diagrams—a Survey of a Fundamental Geometric Data Structure

Franz Aurenhammer. “V oronoi Diagrams—a Survey of a Fundamental Geometric Data Structure”. In:ACM Computing Surveys 23.3 (Sept. 1991), pp. 345–405. ISSN : 0360-0300, 1557-7341. DOI: 10.1145/116873. 116880

1991 doi
[27]

Real-Time Dynamics of Soft and Continuum Robots Based on Cosserat Rod Models

John Till, Vincent Aloi, and Caleb Rucker. “Real-Time Dynamics of Soft and Continuum Robots Based on Cosserat Rod Models”. In: The International Journal of Robotics Research 38.6 (May 2019), pp. 723–746. ISSN : 0278-3649, 1741-3176. DOI: 10.1177/0278364919842269. 16 A PREPRINT...

2019 doi
[28]

Finite Element Modeling of Soft Fluidic Actuators: Overview and Recent Developments

Matheus S. Xavier, Andrew J. Fleming, and Yuen K. Yong. “Finite Element Modeling of Soft Fluidic Actuators: Overview and Recent Developments”. In: Advanced Intelligent Systems 3.2 (Feb. 2021), p. 2000187. ISSN : 2640-4567, 2640-4567. DOI: 10.1002/aisy.202000187

2021 doi
[29]

Topological Optimum Design Using Genetic Algorithms

Couro Kane and Marc Schoenauer. “Topological Optimum Design Using Genetic Algorithms”. In: Control and Cybernetics 25.5 (1996), pp. 1059–1088

1996
[30]

Genetic-Based EM Algorithm for Learning Gaussian Mixture Models

F. Pernkopf and D. Bouchaffra. “Genetic-Based EM Algorithm for Learning Gaussian Mixture Models”. In:IEEE Transactions on Pattern Analysis and Machine Intelligence27.8 (Aug. 2005), pp. 1344–1348. ISSN : 0162-8828. DOI: 10.1109/TPAMI.2005.162. 17

2005 doi
[2884]

DOI: 10.1109/TMECH.2019.2943418

ISSN : 1083-4435, 1941-014X. DOI: 10.1109/TMECH.2019.2943418

1941

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Design, Fabrication and Control of Soft Robots

Daniela Rus and Michael T. Tolley. “Design, Fabrication and Control of Soft Robots”. In: Nature 521.7553 (May 28, 2015), pp. 467–475. ISSN : 0028-0836, 1476-4687. DOI: 10.1038/nature14543

doi:10.1038/nature14543 2015

[2] [2]

Challenges and Opportunities for Design, Simulation, and Fabrication of Soft Robots

Hod Lipson. “Challenges and Opportunities for Design, Simulation, and Fabrication of Soft Robots”. In: Soft Robotics 1.1 (Mar. 2014), pp. 21–27. ISSN : 2169-5172, 2169-5180. DOI: 10.1089/soro.2013.0007

arXiv 2014

[3] [3]

Soft Actuators for Soft Robotic Applications: A Review

Nazek El-Atab, Rishabh B. Mishra, Fhad Al-Modaf, Lana Joharji, Aljohara A. Alsharif, Haneen Alamoudi, Marlon Diaz, Nadeem Qaiser, and Muhammad Mustafa Hussain. “Soft Actuators for Soft Robotic Applications: A Review”. In: Advanced Intelligent Systems 2.10 (Oct. 2020), p. 2000128. ISSN : 2640-4567, 2640-4567. DOI: 10.1002/aisy.202000128

doi:10.1002/aisy.202000128 2020

[4] [4]

From Bioinspiration to Computer Generation: Developments in Autonomous Soft Robot Design

Joshua Pinskier and David Howard. “From Bioinspiration to Computer Generation: Developments in Autonomous Soft Robot Design”. In: Advanced Intelligent Systems 4.1 (Jan. 2022), p. 2100086. ISSN : 2640-4567, 2640-4567. DOI: 10.1002/aisy.202100086

work page doi:10.1002/aisy.202100086 2022

[5] [5]

Evolutionary Black-Box Topology Optimization: Challenges and Promises

David Guirguis, Nikola Aulig, Renato Picelli, Bo Zhu, Yuqing Zhou, William Vicente, Francesco Iorio, Markus Olhofer, Wojciech Matusik, Carlos Artemio Coello Coello, and Kazuhiro Saitou. “Evolutionary Black-Box Topology Optimization: Challenges and Promises”. In: IEEE Transactions on Evolutionary Computation 24.4 (Aug. 2020), pp. 613–633. ISSN : 1089-778X,...

work page arXiv 2020

[6] [6]

Design and Control of Soft Robots Using Dif- ferentiable Simulation

Moritz Bächer, Espen Knoop, and Christian Schumacher. “Design and Control of Soft Robots Using Dif- ferentiable Simulation”. In: Current Robotics Reports 2.2 (June 2021), pp. 211–221. ISSN : 2662-4087. DOI: 10.1007/s43154-021-00052-7

work page doi:10.1007/s43154-021-00052-7 2021

[7] [7]

Towards Automatic Design of Soft Pneumatic Actuators: Inner Structure Design Using CNN Model and Bézier Curve-Based Genetic Algorithm

Loïc Mosser, Laurent Barbé, Lennart Rubbert, and Pierre Renaud. “Towards Automatic Design of Soft Pneumatic Actuators: Inner Structure Design Using CNN Model and Bézier Curve-Based Genetic Algorithm”. In: IEEE Robotics and Automation Letters 8.10 (Oct. 2023), pp. 6603–6610. ISSN : 2377-3766, 2377-3774. DOI: 10.1109/ LRA.2023.3309135

arXiv 2023

[8] [8]

Design and Development of a Soft Gripper with Topology Optimization

Hongying Zhang, Michael Yu Wang, Feifei Chen, Yiqiang Wang, A. Senthil Kumar, and Jerry Y . H. Fuh. “Design and Development of a Soft Gripper with Topology Optimization”. In: 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Vancouver, BC: IEEE,...

work page arXiv 2017