Multi-objective Evolution of Drone Morphology

Christophe de Wagter; Elijah H. W. Ang; Guido C. H. E. de Croon

REVIEW 4 major objections 5 minor 16 references

Multi-objective Evolution of Drone Morphology

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that an evolutionary algorithm, NSGA-II, can design multicopter bodies that beat the standard 5-inch 220 mm quadcopter on at least one of three objectives: the most maneuverable evolved drone improves maneuverability by…

desk verdict Simulation-only but honest evolutionary drone design study; the headline maneuverability metric has a real saturation-margin problem, but the trade-off data and pipeline merit a serious referee. read the letter →

arxiv 2502.01197 v1 pith:QPC7TX25 submitted 2025-02-03 cs.RO

Elijah H. W. Ang , Christophe De Wagter , Guido C. H. E. de Croon This is my paper · ORCID

classification cs.RO

keywords dronemorphologymulticopterdesignNSGA-IImulti-objectiveoptimizationParetofrontthrust-to-weightratiomaneuverabilityevolutionary

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

The paper asks whether an evolutionary search, rather than human intuition, can improve on the classic quadcopter layout. It evolves hundreds of multicopter body plans under three objectives—thrust-to-weight ratio, maneuverability, and size—using the NSGA-II genetic algorithm with a population of 600 evolved over 2000 generations. The best evolved designs beat a standard 5-inch, 220 mm quadcopter in at least one objective: the most maneuverable is 487.8% more maneuverable, the highest-thrust is 23.5% better in thrust-to-weight, and the smallest is 4.8% smaller. The paper argues the maneuverability gain comes from tilted propellers and smaller size, while the thrust gain comes from extra propellers. It concludes that the standard quadcopter remains on the Pareto front, but it is not the only optimal answer.

What carries the argument

The machinery is the actuator effectiveness matrices $B_f$ and $B_m$, which encode how each propeller's position, tilt, azimuth, and spin direction contribute to net force and torque. From $B_m$ the paper forms the controllability Gramian $W_c = B_m B_m^T$; the smallest eigenvalue of this matrix becomes the maneuverability score. Thrust-to-weight is computed by first finding hovering control inputs that minimize actuator effort, then scaling until one actuator saturates. NSGA-II is the non-dominated sorting genetic algorithm that explores the genotype space—number of propellers, arm lengths, arm angles, and propeller orientations—and returns a Pareto front of non-dominated designs.

What would settle it

Build physical copies of the highest-thrust and most-maneuverable evolved designs plus the standard 220 mm quadcopter, measure thrust-to-weight on a test stand and angular acceleration in yaw, and check whether the 487.8% and 23.5% margins appear. If they shrink to noise, the linear actuator model or the Gramian maneuverability metric is the broken link.

Watch

Extended reading notes

Core claim

The central discovery is that the design space of multicopter morphology contains solutions that outperform the standard quadcopter by large margins on single objectives, but no single evolved morphology dominates all three objectives simultaneously. Specifically, evolution found a drone with 487.8% higher maneuverability—measured as the smallest eigenvalue of the controllability Gramian $B_m B_m^T$—a drone with 23.5% higher thrust-to-weight ratio, and a drone 4.8% smaller than the 220 mm quadcopter baseline. These three extrema are Pareto-optimal, meaning no other evolved design is better in all three objectives at once. The paper attributes the maneuverability improvement to propeller tilt and compact layout, and the thrust improvement to a higher number of propellers, while noting that the standard quadcopter itself lies on the Pareto front and is not dominated.

Load-bearing premise

The results hinge on the assumption that the simulation—constant propeller coefficients, linearized actuator commands, and the specified battery-and-controller mass model—predicts real drone behavior closely enough that the simulated gains appear on physical hardware, and that the smallest eigenvalue of the moment matrix is a true measure of maneuverability.

Editorial extensions

If this is right

If the model predictions hold on hardware, drone designers can use evolutionary search to generate mission-specific body plans instead of starting from a standard quadcopter layout.
Tilted-propeller layouts appear to be a practical route to much higher yaw maneuverability, which is relevant for drone racing and physical interaction tasks.
Increasing propeller count is the effective lever for thrust-to-weight ratio, but it forces larger size; the Pareto front quantifies that trade-off.
The standard quadcopter's position on the Pareto front means it is a reasonable default design, but for missions that emphasize one objective, other evolved morphologies are better.
The same fitness-evaluation approach can incorporate mission-specific constraints such as payload placement or sensor fields of view.

Reading between the lines

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

Editorial inference: If the model transfers to hardware, the 487.8% maneuverability number is likely metric-specific; a physical yaw-step test would probably show a smaller but still substantial gain.
Editorial inference: The simulation fixes propeller size and uses constant propeller coefficients, so the most tilted or octocopter designs may suffer aerodynamic interference in reality; a thrust-stand comparison would reveal how much.
Editorial inference: The same fitness evaluation could be used for co-design of morphology with controller tuning or payload placement, where the Pareto front would differ from the one reported here.
Editorial inference: Because the standard quadcopter sits on the Pareto front, the practical message is to treat it as a reference point in a broader search rather than as a design to discard.

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 genuinely new thing here is the end-to-end pipeline: NSGA-II over a parameterized multicopter morphology (propeller count, arm length and angle, tilt, azimuth, spin direction) with fast model-based fitness that checks static and spinning hover feasibility, then scores thrust-to-weight, a controllability-Gramian maneuverability index, and size. That combination is new in the drone design literature, which has mostly stayed with convex or single-objective optimization. The paper is clearly written and the authors are upfront that results are simulation-only, with flight tests deferred to future work. The Pareto-front analysis gives useful, concrete intuition about how propeller count and tilt trade against the three objectives. Credit where due: the non-static hover condition (spinning hover) is a nice touch that keeps near-viable designs in the gene pool, and the genotype with phantom propellers is a pragmatic trick for fixed-length encoding with variable propeller count. The soft spots are real but not fatal. The stress-test note about the maneuverability metric is correct: min eigenvalue of B_m B_m^T treats actuator commands as a symmetric centered unit ball, while actual inputs are one-sided with a hover-dependent saturation margin. So the 487.8% maneuverability improvement over the quadcopter is not yet supported as a physical claim; it is an artifact of an unvalidated proxy. A better metric would compute worst-axis angular acceleration over the actual box [-eta_hat, 1-eta_hat], which would also reward hover-point margins. The thrust-to-weight and size results are less affected, since those derive directly from the static hover solution. Also, the paper reports a single NSGA-II run with no repeated seeds and no code, data, or full parameter values; that matters for a stochastic optimizer. These are important omissions for reproducibility, but they are omissions rather than fundamental errors. The physics model is standard and the paper does not fit parameters to its headline results, so the circularity burden is low. Bottom line: this is a promising methodology paper for drone engineers and evolutionary design researchers, not a definitive empirical demonstration. It deserves peer review, but a referee should press for the corrected maneuverability metric, repeated runs, and a commitment to release code and data. I would take it to review, and cite it once the metric is fixed.

Referee Report

4 major / 5 minor

Summary. The paper proposes an evolutionary multi-objective optimization framework for multicopter morphology design. The authors model drones via actuator effectiveness matrices B_f and B_m, compute a minimum-energy hover solution (including a spinning-hover fallback), and define three objectives: thrust-to-weight ratio from a scaled hover solution (Eqs. 5-6), maneuverability as the smallest eigenvalue of the moment controllability Gramian (Eqs. 7-8), and size as the convex hull volume of the propeller positions (Sec. III.F). Using NSGA-II with a population of 600 evolved over 2000 generations, they report a Pareto front of designs that outperform a standard 220 mm quadcopter in at least one of the three objectives, with improvements of 487.8% in maneuverability, 23.5% in thrust-to-weight, and 4.8% in size. They attribute the maneuverability gain to tilted propellers and the thrust gain to additional propellers.

Significance. The paper addresses a timely and interesting problem: automated design of multicopter morphology. The model-based fitness pipeline is computationally efficient, the genotype-phenotype mapping includes practical constraints such as propeller spacing, and the use of NSGA-II with a clear baseline comparison is appropriate. The main conceptual contribution, treating morphology as an evolvable design variable with a multi-objective fitness landscape, could be valuable for design-space exploration. However, the quantitative claims are not yet credible as physical performance improvements because the maneuverability metric ignores actuator saturation and one-sided input constraints, the thrust-to-weight metric is a feasible lower bound rather than a true maximum, and the model is unvalidated and incompletely specified. The identified issues are addressable within the manuscript's scope, so the paper merits major revision rather than rejection.

major comments (4)

[Sec. V.B, Eqs. (7)-(8)] The maneuverability objective is the smallest eigenvalue of W_c = B_m B_m^T, the controllability Gramian for unbounded symmetric inputs. The actual actuator commands are one-sided, η ∈ [0,1]^n (Eq. 2), and the available deviations around the hover point η̂ are δ ∈ [-η̂, 1-η̂]. The reachable angular-acceleration set is a translated zonotope, not the centered ellipsoid B_m B_m^T. A design with large B_m entries may hover near saturation and have poor real worst-axis authority, while a design with smaller B_m but larger margins may be better. Thus the 487.8% improvement and the attribution to tilt angles are not supported as physical maneuverability. Please recompute the metric from the actual feasible box, e.g., the minimum over directions of the maximum achievable angular acceleration subject to δ ∈ [-η̂, 1-η̂].
[Sec. III.D, Eq. (5)] The paper computes the thrust-to-weight ratio by uniformly scaling the minimum-energy hover solution until one input saturates. This yields a feasible point, but the true maximum of ||B_f η||/g over {η ∈ [0,1]^n : B_m η = 0} may occur at a different point, since the optimization is non-convex and the hover direction need not remain optimal as inputs saturate. Therefore the label "maximum possible thrust-to-weight ratio" is inaccurate, and the evolutionary ranking uses a proxy that may differ from the actual maximum. Please either solve the true allocation problem or explicitly describe the value as a feasible lower bound.
[Sec. IV.B and V] The results are based on a single NSGA-II run with a population of 600 over 2000 generations, but the paper provides no information on random seeds, number of runs, or variability of the Pareto front. Because the algorithm is stochastic, the specific improvements (487.8%, 23.5%, 4.8%) and the elimination of 8-propeller designs may be run-specific. Please add repeated runs with variance or a sensitivity analysis to support the quantitative claims.
[Sec. III and V.C] The simulation model is not fully specified for reproducibility. The values of k_f, k_m, m, μ, ω_max, the propeller mass, and the inertia model for the flight-controller stack are not given; the mass/inertia model is described only verbally. The paper also acknowledges that all results are from simulation with flight tests deferred to future work. Without these parameters and model details, the central quantitative claims cannot be reproduced or assessed, and the expected "reality gap" is unknown. Please include a complete parameter table and, if possible, a basic validation against a known quadcopter.

minor comments (5)

[Sec. III.A, Eq. (2)] The definitions of B_f and B_m are not written out; please give the explicit expressions in terms of propeller positions, orientations, and inertia.
[Sec. IV.A] The parameter values for the propeller coefficients, motor mass, and arm density are absent; a table of all simulation parameters would greatly improve reproducibility.
[Sec. V.B] The statement that "size, maneuverability and thrust-to-weight ratio are very clearly clustered" is based on visual inspection of Figures 4-6; consider adding a quantitative clustering or Pareto analysis.
[Sec. V.C, Eq. (9)] Equation (9) is introduced without derivation or reference; please define m, μ, and I_fc precisely and explain how the angular acceleration depends on arm length.
[Sec. VI] The conclusion repeats the claim of "well-studied drone physics" despite the acknowledged reality gap; consider softening this to reflect the unvalidated model.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline numbers are the optimized values of the paper's own fixed model-based objectives, with no parameter fitted to the result and no load-bearing self-citation.

full rationale

The evolution pipeline is self-contained: the genotype parameters (arm lengths, angles, tilts, propeller counts) are inputs, and the three objectives—thrust-to-weight ratio (Eqs. 5–6), maneuverability (Eqs. 7–8), and size (Section III.F)—are fixed functions of the morphology. NSGA-II maximizes/minimizes these functions, and the reported 487.8%, 23.5%, and 4.8% improvements are exactly the objective values of the Pareto-extreme individuals compared with a fixed 220 mm quadcopter baseline under the same model. No parameter is fitted to the reported improvements, and no 'prediction' is derived from the quantity it is supposed to predict; the optimization result is the output, not an input. There are no self-citations in the reference list, and the Gramian and NSGA-II are standard external tools [14,16]. The paper explicitly limits its claims to simulation ('Currently, the results from the evolution are evaluated in simulation, with the manufacturing and flight tests of prototypes to be performed as part of our future work'), so concerns about the actuator-effectiveness or inertia model are matters of model validity, not circularity. The derivation chain therefore does not reduce to its own inputs.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim depends on several unstated model constants and ad hoc objective definitions. The quantitative improvements are only as good as these assumptions.

free parameters (5)

Propeller thrust coefficient k_f = not specified in paper
Absent from the paper; sets the scale of thrust and TWR in Eq. (1)-(6).
Propeller moment coefficient k_m = not specified in paper
Absent; controls moment generation and the maneuverability Gramian in Eq. (7).
Motor mass m = not specified in paper
Used in Eq. (9) for angular acceleration and in inertia computations; not given.
Arm mass per unit length mu = not specified in paper
Used in Eq. (9) for arm inertia; not given.
Maximum propeller speed omega_max = not specified in paper
Combined with k_f and k_m to form actuator effectiveness matrices; not given.

assumptions (5)

domain assumption Thrust and moment are proportional to the square of propeller speed (Eq. 1)
Standard actuator model for propellers, but unverified for the specific propellers used.
ad hoc to paper Maneuverability is measured by the smallest eigenvalue of B_m B_m^T (Eq. 7-8)
A chosen proxy; no justification that it predicts real agility.
ad hoc to paper The energy-minimizing hover solution can be scaled uniformly to reach the maximum thrust-to-weight ratio (Eq. 5)
This is asserted; a different control allocation could yield a larger TWR.
ad hoc to paper Spinning hover (Eq. 4) is a viable fallback for non-static configurations
The constraint ||B_f eta x B_m eta||=0 is introduced so spinning drones count as viable, but physical feasibility is not demonstrated.
domain assumption All motors lie in the x-y plane and the flight controller stack is a fixed 250 g
Design constraint; limits the search space and affects the Pareto front.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Multi-objective Evolution of Drone Morphology." pith.science (2026). https://pith.science/paper/QPC7TX25

@misc{pith2026250201197,
  author       = {Pith},
  title        = {Pith review of: Multi-objective Evolution of Drone Morphology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QPC7TX25}},
  note         = {Machine review of arXiv:2502.01197}
}

read the original abstract

The design of multicopter drones has remained almost the same since its inception. While conventional designs, such as the quadcopter, work well in many cases, they may not be optimal in specific environments or missions. This paper revisits rotary drone design by exploring which body morphologies are optimal for different objectives and constraints. Specifically, an evolutionary algorithm is used to produce optimal drone morphologies for three objectives: (1) high thrust-to-weight ratio, (2) high maneuverability, and (3) small size. To generate a range of optimal drones with performance trade-offs between them, the non-dominated sorting genetic algorithm II, or NSGA-II is used. A randomly sampled population of 600 is evolved over 2000 generations. The NSGA-II algorithm evolved drone bodies that outperform a standard 5-inch 220 mm wheelbase quadcopter in at least one of the three objectives. The three extrema in the Pareto front show improvement of 487.8%, 23.5% and 4.8% in maneuverability, thrust-to-weight ratio and size, respectively. The improvement in maneuverability can be attributed to the tilt angles of the propellers, while the increase in thrust-to-weight ratio is primarily due to the higher number of propellers. The quadcopter is located on the Pareto front for the three objectives. However, our results also show that other designs can be better depending on the objectives.

Figures

Figures reproduced from arXiv: 2502.01197 by the authors.

**Figure 2.** Illustration of genotype used for evolution. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 4.** Scatter plot of Pareto optimal drones’ thrust-to-we [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Figures from the paper (3 more)

**Figure 6.** Figure 6: Scatter plot of Pareto optimal drones’ thrust-to-we [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

**Figure 7.** Figure 7: Drone with highest thrust-to-weight ratio after gen [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

**Figure 8.** Figure 8: Smallest and most maneuverable drone after generati [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 15 canonical work pages

[1]

Classiﬁcations, appli cations, and design challenges of drones: A review,

M. Hassanalian and A. Abdelkeﬁ, “Classiﬁcations, appli cations, and design challenges of drones: A review,” Progress in Aerospace sci- ences, vol. 91, pp. 99–131, 2017

work page 2017
[2]

A survey on design and development o f an unmanned aerial vehicle (quadcopter),

D. BBVL and P . Singh, “A survey on design and development o f an unmanned aerial vehicle (quadcopter),” International Journal of Intelligent Unmanned Systems , vol. 4, no. 2, pp. 70–106, 2016

work page 2016
[3]

Development, implementation, and exp erimen- tal outdoor evaluation of quadcopter controllers for compu tationally limited embedded systems,

J. Paredes, P . Sharma, B. Ha, M. Lanchares, E. Atkins, P . G askell, and I. Kolmanovsky, “Development, implementation, and exp erimen- tal outdoor evaluation of quadcopter controllers for compu tationally limited embedded systems,” Annual Reviews in Control , vol. 52, pp. 372–389, 2021

work page 2021
[4]

Evolving robot morp hology,

H. H. Lund, J. Hallam, and W.-P . Lee, “Evolving robot morp hology,” in Proceedings of 1997 IEEE International Conference on Evolu tionary Computation (ICEC’97) . IEEE, 1997, pp. 197–202

work page 1997
[5]

Genetic algorithms: principles of natural selection applied to computation,

S. Forrest, “Genetic algorithms: principles of natural selection applied to computation,” Science, vol. 261, no. 5123, pp. 872–878, 1993

work page 1993
[6]

Computer-auto mated evolution of an X-band antenna for NASA’s space technology 5 mission,

G. S. Hornby, J. D. Lohn, and D. S. Linden, “Computer-auto mated evolution of an X-band antenna for NASA’s space technology 5 mission,” Evolutionary computation , vol. 19, no. 1, pp. 1–23, 2011

work page 2011
[7]

Evolut ionary de- velopmental robotics: Improving morphology and control of physical robots,

V . Vujovic, A. Rosendo, L. Brodbeck, and F. Iida, “Evolut ionary de- velopmental robotics: Improving morphology and control of physical robots,” Artiﬁcial life , vol. 23, no. 2, pp. 169–185, 2017

work page 2017
[8]

Optimizing modular robot compos ition: A lexicographic genetic algorithm approach,

J. Külz and M. Althoff, “Optimizing modular robot compos ition: A lexicographic genetic algorithm approach,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 16 752–16 758

work page 2024

Show all 16 references

[9]

Co-design optimisation of morphing topol ogy and control of winged drones,

F. Bergonti, G. Nava, V . Wüest, A. Paolino, G. L’Erario, D . Pucci, and D. Floreano, “Co-design optimisation of morphing topol ogy and control of winged drones,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 8679–8685

2024
[10]

Evolutiona ry algorithms for multi-objective optimization of drone controller para meters,

A. Shamshirgaran, H. Javidi, and D. Simon, “Evolutiona ry algorithms for multi-objective optimization of drone controller para meters,” in 2021 IEEE Conference on Control Technology and Application s (CCTA). IEEE, 2021, pp. 1049–1055

2021
[11]

Position control of a quadcopter drone using evolutionary algorithms-based self-tuning fo r ﬁrst-order Takagi–Sugeno–Kang fuzzy logic autopilots,

E. Y azid, M. Garratt, and F. Santoso, “Position control of a quadcopter drone using evolutionary algorithms-based self-tuning fo r ﬁrst-order Takagi–Sugeno–Kang fuzzy logic autopilots,” Applied Soft Comput- ing, vol. 78, pp. 373–392, 2019

2019
[12]

Comp utational multicopter design,

T. Du, A. Schulz, B. Zhu, B. Bickel, and W. Matusik, “Comp utational multicopter design,” ACM Transactions on Graphics , 2016

2016
[13]

Robot co-design: beyond t he monotone case,

L. Carlone and C. Pinciroli, “Robot co-design: beyond t he monotone case,” in 2019 International Conference on Robotics and Automation (ICRA). IEEE, 2019, pp. 3024–3030

2019
[14]

S. L. Brunton and J. N. Kutz, Data-driven science and engineering: Machine learning, dynamical systems, and control . Cambridge University Press, 2022

2022
[15]

Design of multirotor aerial vehicles: A taxonomy based on i nput allocation,

M. Hamandi, F. Usai, Q. Sablé, N. Staub, M. Tognon, and A. Franchi, “Design of multirotor aerial vehicles: A taxonomy based on i nput allocation,” The International Journal of Robotics Research , vol. 40, no. 8-9, pp. 1015–1044, 2021

2021
[16]

A fast a nd elitist multiobjective genetic algorithm: NSGA-II,

K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast a nd elitist multiobjective genetic algorithm: NSGA-II,” IEEE transactions on evolutionary computation, vol. 6, no. 2, pp. 182–197, 2002

2002

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Classiﬁcations, appli cations, and design challenges of drones: A review,

M. Hassanalian and A. Abdelkeﬁ, “Classiﬁcations, appli cations, and design challenges of drones: A review,” Progress in Aerospace sci- ences, vol. 91, pp. 99–131, 2017

work page 2017

[2] [2]

A survey on design and development o f an unmanned aerial vehicle (quadcopter),

D. BBVL and P . Singh, “A survey on design and development o f an unmanned aerial vehicle (quadcopter),” International Journal of Intelligent Unmanned Systems , vol. 4, no. 2, pp. 70–106, 2016

work page 2016

[3] [3]

Development, implementation, and exp erimen- tal outdoor evaluation of quadcopter controllers for compu tationally limited embedded systems,

J. Paredes, P . Sharma, B. Ha, M. Lanchares, E. Atkins, P . G askell, and I. Kolmanovsky, “Development, implementation, and exp erimen- tal outdoor evaluation of quadcopter controllers for compu tationally limited embedded systems,” Annual Reviews in Control , vol. 52, pp. 372–389, 2021

work page 2021

[4] [4]

Evolving robot morp hology,

H. H. Lund, J. Hallam, and W.-P . Lee, “Evolving robot morp hology,” in Proceedings of 1997 IEEE International Conference on Evolu tionary Computation (ICEC’97) . IEEE, 1997, pp. 197–202

work page 1997

[5] [5]

Genetic algorithms: principles of natural selection applied to computation,

S. Forrest, “Genetic algorithms: principles of natural selection applied to computation,” Science, vol. 261, no. 5123, pp. 872–878, 1993

work page 1993

[6] [6]

Computer-auto mated evolution of an X-band antenna for NASA’s space technology 5 mission,

G. S. Hornby, J. D. Lohn, and D. S. Linden, “Computer-auto mated evolution of an X-band antenna for NASA’s space technology 5 mission,” Evolutionary computation , vol. 19, no. 1, pp. 1–23, 2011

work page 2011

[7] [7]

Evolut ionary de- velopmental robotics: Improving morphology and control of physical robots,

V . Vujovic, A. Rosendo, L. Brodbeck, and F. Iida, “Evolut ionary de- velopmental robotics: Improving morphology and control of physical robots,” Artiﬁcial life , vol. 23, no. 2, pp. 169–185, 2017

work page 2017

[8] [8]

Optimizing modular robot compos ition: A lexicographic genetic algorithm approach,

J. Külz and M. Althoff, “Optimizing modular robot compos ition: A lexicographic genetic algorithm approach,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2024, pp. 16 752–16 758

work page 2024