Pith. sign in

REVIEW 3 major objections 4 minor 87 references

Environmental resilience via morphological diversity within machines

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A simulation study argues that physical connectors, trained to restore motion to tethered pairs of diverse simulated robots, acquire a broad enough repertoire of internal disturbance handling that novel environments fall within their manage

desk verdict A solid empirical demonstration that diversity-trained connectors generalize to novel environments; the mechanism claim is plausible but rests on thin evidence. read the letter →

arxiv 2608.02395 v1 pith:ZYA2JS2D submitted 2026-08-03 cs.RO

Alice Hein , Josh Bongard This is my paper
classification cs.RO
keywords morphologicaldiversitycollectiveresiliencesoftrobotsneuralcontrollersloosecouplingevolutionaryroboticsinternaladversitynewenvironments
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 proposes a mechanism for an old idea: that the diversity present inside organisms helps them survive surprise. In a simulated world, pairs of independently trained soft robots are tethered together by connectors, and those connectors are trained to restore forward motion. When the connectors have been trained on morphologically diverse pairs, the disturbances they learned to tame are broad enough that later encounters with new environments fall within that range, so the collective keeps moving without retraining. The authors find that this resilience grows with the number or diversity of the components. The work suggests that intentionally creating internal physical adversity might prepare both organisms and machines for external adversity.

What carries the argument

The central object is the physical connector: five actuated springs with a shared neural network controller that tether two or more simulated agents. The connector is trained by gradient descent to maximize the forward displacement of the tethered pair, and in doing so it must absorb and compensate for whatever behavioral differences the two agents exhibit. The paper uses a nonlinear dimensionality-reduction visualization (PaCMAP) of the connector's hidden states to argue that states encountered while training on diverse pairs largely overlap with states encountered in novel environments, making the new worlds 'familiar' to a diversity-trained connector.

What would settle it

Measure the actual distribution of disturbances (for example, dynamic time warping distances between tethered and independent behavior) during connector training on diverse pairs and during deployment to each novel environment. If a novel environment's disturbance distribution lies outside the training distribution yet diverse-trained connectors still outperform uniform-trained ones, the proposed overlap mechanism would be falsified. Alternatively, an environment where uniform-trained connectors do better would contradict the claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that physical connectors, in the process of learning to restore rightward locomotion to pairs of previously independent, morphologically diverse agents they have tethered together, trigger and tame a sufficiently diverse set of disruptions that when the same connectors are later placed in novel environments, the disruptions caused by those environments fall within this manageable range. As a result, the whole collective can continue to move properly with no further adaptation. The authors also report that increasing the number of agents or their morphological diversity further increases resilience, and that diversity-trained connectors generalize to uniform colle

Load-bearing premise

The explanation relies on the assumption that the range of disruptions experienced when tethering diverse agents during training overlaps with the range caused by novel environments; the paper's direct evidence is a single PaCMAP visualization of one connector per condition, and its own supplemental table shows the overlap is not higher for diverse-trained connectors in sticky ground, sand, and treadmill.

Editorial extensions

If this is right

  • Collectives built from loosely coupled, morphologically diverse subunits can handle new environments without retraining.
  • Scaling up—using more components or more diverse components—improves this resilience.
  • Connectors trained on diverse pairs also perform well in uniform collectives, suggesting the training transfers across collective structures.
  • The approach offers a possible alternative to training in many environments; internal diversity during training may substitute for external variety.

Reading between the lines

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

  • The authors' overlap mechanism could be tested more directly by measuring the physical disturbances (e.g., forces or positional errors) during training and deployment, rather than only visualizing hidden states.
  • If the mechanism holds, it suggests a concrete design principle for modular robotics: purposely introduce heterogeneity and loose coupling early, rather than standardize all subunits.
  • The idea may generalize beyond mechanical tethering—any process that forces a controller to absorb a wide range of internal perturbations could pre-tune it for external changes.
  • The paper leaves open whether the resilience advantage arises from the connector's learned policy or simply from the diversity of the agents themselves; a controlled comparison could separate these.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper investigates whether morphological diversity within a machine collective, combined with trainable coupling, confers resilience to novel environments. In simulation, agents are evolved to locomote, then trained connectors tether pairs of agents and are optimized to restore collective forward motion. Four connector types are trained (weak/strong connectors, on uniform or diverse agent pairs) and deployed in eight collective types across 12 novel environments. The central empirical finding is that weak connectors trained and deployed within diverse agent pairs lose the least performance when environments change, and that this advantage grows with collective size and with the degree of agent diversity. The authors propose a mechanism: during training on diverse pairs, connectors experience and "tame" a wide range of internal disruptions, so that disruptions caused by novel environments fall within the connectors' manageable range. This mechanism is supported by a PaCMAP-based state-overlap analysis of connector hidden states, though the analysis is limited and not fully consistent across environments.

Significance. If the empirical effect is robust, the paper makes a useful contribution to evolutionary robotics and artificial life: it suggests that controlled internal heterogeneity and loose coupling can serve as a form of pre-adaptation to external novelty, without needing to train on the target environment. The scaling results (more agents or more diversity increases resilience) are a falsifiable prediction and, within the simulation, appear to support the hypothesis. The main value lies in this principle rather than in the specific simulations. However, the paper's headline claim is explicitly mechanistic, and the evidence for the proposed causal mechanism is presently under-supported; this limits the confidence with which the principle can be generalized.

major comments (3)
  1. [§3.3, Fig. 2B] The transplant experiment is the central demonstration that diversity-trained connectors generalize to uniform collectives and that uniformity-trained connectors fail to exploit diversity. Fig. 2B shows four bars with no error bars, no confidence intervals, and no statistical tests. Given that each data point aggregates 10 connectors and 100 agent pairs, per-connector variation could be substantial. Please provide the underlying distributions and paired statistical comparisons for the two transplant conditions.
  2. [§4.1, Fig. 3 and Table S1] The proposed mechanism—that resilience arises because novel environments produce connector hidden states overlapping with training states—is not consistently supported by Table S1. In sticky ground, sand, and treadmill, diverse-trained connectors do not show higher overlap than uniformity-trained connectors, and 'present only in training' counts are not consistently lower for diverse training. Since the mechanism is claimed to explain the resilience advantage, the three negative environments need discussion or a refined overlap measure. Additionally, the overlap metric uses PaCMAP-reduced 2D embeddings rounded to integer coordinates; it is not shown that this coarse reduction preserves the 32-dimensional state-space relationships that determine connector behavior. A direct distance-based overlap or a validation of the embedding's fidelity would strengthen the claim.
  3. [§3.3 and §4.4, Fig. 2A] The resilience advantage is demonstrated only for weak connectors. Strong connectors, when deployed in novel environments, perform poorly across all collective types (Fig. 2A, rightmost four small bars). The abstract and title refer to 'physical connectors' and 'machines' without this qualifier. This is a significant boundary condition that should be stated explicitly in the abstract or at least in the conclusions, and the mechanistic discussion should explain why stiffness eliminates the effect.
minor comments (4)
  1. [Fig. 2A caption] "Error bars are visually unresolvable" is not a substitute for reporting uncertainty. Please provide exact values and confidence intervals in the supplement or in the figure itself.
  2. [Table S1] The table reports raw counts without statistical tests or effect sizes. A simple paired test across the 10 connector models would help assess whether the overlap differences are meaningful beyond the three highlighted cases.
  3. [§7.1.2, §7.1.4] The supplementary equations are numbered inconsistently (e.g., '1.2.1', '1.4.1'). Please use a single consistent numbering scheme.
  4. [§4.1, Fig. 3B,C] The overlap panel shows one representative connector per condition. Please state explicitly how representative these two connectors are and how many of the 10 models per condition show the displayed pattern.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central resilience result is an independent empirical measurement and the proposed overlap mechanism is post hoc, not definitionally tied to the outcome.

full rationale

The paper's derivation chain is self-contained. Connectors are trained only to restore forward motion of tethered agent pairs on flat ground (Section 2.2), and the headline resilience numbers are measured by deploying the resulting, frozen controllers into 12 environments not used for training (Sections 2.3-3.3). No parameter is fitted to those deployment outcomes and then relabeled as a prediction; the diverse-vs-uniform comparison is a direct behavioral measurement. The proposed mechanism in Section 4.1 is an after-the-fact PaCMAP overlap analysis of 32-D connector hidden states, and Table S1 shows the overlap advantage for diverse-trained connectors is absent in sticky ground, sand, and treadmill. That weakens the causal explanation, but it is not circular: overlap and resilience are distinct measured quantities, and overlap is not constructed from the resilience values. The only notable author self-citation is [84] (Matthews et al., including Bongard), used to justify the alternating evolutionary/gradient-descent component optimizer ('This alternation follows 84 who showed...'); the method is fully described and implemented here, and the paper's central resilience claim does not reduce to that citation. No equation in the paper equates an input to the output it is claimed to predict, so no circular step is present.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

All quantities are simulation-based; no fitted equations or invented entities. The free parameters are hand-chosen stiffness levels that define the 'loose coupling' condition, and the axioms concern the representativeness of the environments and the post-hoc state-space analysis.

free parameters (2)
  • weak connector spring stiffness = not reported (categorical low stiffness)
    The main resilience result is demonstrated with weak connectors; the effect may not hold for other stiffness values, and no sensitivity sweep is provided.
  • strong connector spring stiffness = not reported (categorical high stiffness)
    Strong connectors performed poorly and were excluded from further analysis; the choice of stiffness splits the results.
assumptions (3)
  • domain assumption The 12 simulated novel environments are representative of the space of challenges a collective might face
    The conclusion that diversity-trained connectors confer general resilience to 'new environments' rests on this set; no sampling or guarantee of representativeness is given.
  • ad hoc to paper PaCMAP embeddings of connector hidden states preserve the similarity structure relevant to the connector's ability to act
    The overlap analysis in Fig 3B/C and Table S1 is used as evidence for the mechanism; this is a projection to 2D with rounding, and the causal connection to behavior is assumed.
  • domain assumption The morphological diversity produced by the evolutionary algorithm is analogous to biological internal diversity
    The biological implications in Sec 4.3 depend on this analogy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Environmental resilience via morphological diversity within machines." pith.science (2026). https://pith.science/paper/ZYA2JS2D

@misc{pith2026260802395,
  author       = {Pith},
  title        = {Pith review of: Environmental resilience via morphological diversity within machines},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZYA2JS2D}},
  note         = {Machine review of arXiv:2608.02395}
}
read the original abstract

Organisms contain diverse, sensorimotor parts across size scales and rapidly adapt to new environments, while machines contain only inert materials at smaller scales and struggle with surprise. We hypothesize that this agents-within-agents quality of organisms may aid their resilience: increasing experiences with internal physical adversity may pre-train organisms and machines to handle external adversity, such as encounters with new environments. Not only has this hypothesis not yet been articulated, mechanisms enabling this phenomenon have yet to be proposed. Here we show a mechanism by which this can occur: we found that physical connectors, in learning to restore behavior to previously independent, morphologically diverse agents they disrupted by tethering them together, trigger and tame sufficiently diverse disruptions that later encounters with new environments trigger disruptions that fall within this manageable range, enabling the collective to continue behaving properly without any additional learning or adaptation. Further, we found that building collectives from more agents, or more diverse agents, further increases the collective's resilience to new environments. This suggests that not just taming but intentionally creating internal physical adversity may indeed prepare organisms for external adversity, and could do so for machines, if they were built from smaller machines.

Figures

Figures reproduced from arXiv: 2608.02395 by the authors.

Figure 1
Figure 1. Internal diversity across scales in nature. A cell’s (a), anatomical joint’s (b), and organism’s (c; blue) experiences are broadened by diverse cell types, changing relative body part sizes, and different caste members encountered during cell migration, development, and locomotion, respectively. This may help damaged cells, visual inversion via prismatic glasses, or a dangling conspecific (red) fall within the cell’… view at source ↗
Figure 2
Figure 2. Internal diversity and loose coupling confer resilience. a, Mean forward motion of independent agents trained on flat ground (dashed dotted) and when deployed to 12 new environments (dotted). Mean forward travel achieved by weak connectors (left two large bars) trained within uniform (orange) or diverse (blue) collectives and strong connectors (right two large bars) trained within uniform (orange) or diverse (blue) … view at source ↗
Figure 3
Figure 3. Internal adversity. a, Connectors within diverse collectives experienced greater internal adversity than those within uniform collectives: Tethered agents within diverse col￾lectives (blue) were disrupted more than agent clones within uniform collectives (orange). (Mean dynamic time warping distance (DTWD) was used to quantify how much a tethered agent’s behavior differed from its independent behavior, after the con… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Performance of uniform and diverse collectives across environments and collective sizes. Orange dashed lines show results for weak connectors trained within uniform two-agent collectives but deployed within larger, uniform collectives. Blue lines show results for weak …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

87 extracted references · 1 linked inside Pith

  1. [1]

    D., Lihoreau, M

    Leonhardt, S. D., Lihoreau, M. & Spaethe, J. Mechanisms of nutritional resource exploitation by insects (2020)

    2020

  2. [2]

    Biological robustness.Nature Reviews Genetics5, 826–837 (2004)

    Kitano, H. Biological robustness.Nature Reviews Genetics5, 826–837 (2004)

    2004

  3. [3]

    InProceedings of the Conference of 16 the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL), 1819–1862 (2024)

    Wu, Z.et al.Reasoning or reciting? Exploring the capabilities and limitations of language models through counterfactual tasks. InProceedings of the Conference of 16 the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL), 1819–1862 (2024)

    2024

  4. [4]

    Gawlikowski, J.et al.A survey of uncertainty in deep neural networks.Artificial Intelligence Review56, 1513–1589 (2023)

    2023

  5. [5]

    Bjerknes, J. D. & Winfield, A. F. T.On fault tolerance and scalability of swarm robotic systems, 431–444 (Springer Berlin Heidelberg, 2013)

    2013

  6. [6]

    Heaven, D.et al.Why deep-learning AIs are so easy to fool.Nature574, 163–166 (2019)

    2019

  7. [7]

    Tajfel, H.Human groups and social categories: Studies in social psychology(Cambridge University Press, Cambridge London New York, 1981)

    1981

  8. [8]

    & Ben-Jacob, E

    Ariel, G., Rimer, O. & Ben-Jacob, E. Order–disorder phase transition in heteroge- neous populations of self-propelled particles.Journal of Statistical Physics158, 579–588 (2015)

    2015

Show all 87 references
  1. [9]

    J., Mirasso, C

    Tessone, C. J., Mirasso, C. R., Toral, R. & Gunton, J. D. Diversity-induced resonance. Physical Review Letters97, 194101 (2006)

    2006

  2. [10]

    Altschuler, S. J. & Wu, L. F. Cellular heterogeneity: Do differences make a difference? Cell141, 559–563 (2010)

    2010

  3. [11]

    Generation of individual diversity: A too neglected fundamental property of adaptive immune system.Frontiers in Immunology5, 208 (2014)

    Muraille, E. Generation of individual diversity: A too neglected fundamental property of adaptive immune system.Frontiers in Immunology5, 208 (2014)

    2014

  4. [12]

    Szwaykowska, K., Romero, L. M.-y.-T. & Schwartz, I. B. Collective motions of het- erogeneous swarms.IEEE Transactions on Automation Science and Engineering12, 810–818 (2015)

    2015

  5. [13]

    Q., Merrin, J., Chait, R., Kowalik, L

    Balaban, N. Q., Merrin, J., Chait, R., Kowalik, L. & Leibler, S. Bacterial persistence as a phenotypic switch.Science305, 1622–1625 (2004)

    2004

  6. [14]

    Kearns, D. B. & Losick, R. Cell population heterogeneity during growth of bacillus subtilis.Genes & Development19, 3083–3094 (2005). 17

    2005

  7. [15]

    Peng, H.et al.Morphological diversity of single neurons in molecularly defined cell types.Nature598, 174–181 (2021)

    2021

  8. [16]

    & Knebel, D.Variability and heterogeneity in natural swarms: Experiments and modeling, 1–33 (Springer International Publishing, Cham, 2022)

    Ariel, G., Ayali, A., Be’er, A. & Knebel, D.Variability and heterogeneity in natural swarms: Experiments and modeling, 1–33 (Springer International Publishing, Cham, 2022)

    2022

  9. [17]

    & Ayali, A

    Knebel, D., Sha-Ked, C., Agmon, N., Ariel, G. & Ayali, A. Collective motion as a distinct behavioral state of the individual.iScience24(2021)

    2021

  10. [18]

    & Gamfeldt, L

    Roger, F., Godhe, A. & Gamfeldt, L. Genetic diversity and ecosystem functioning in the face of multiple stressors.PLOS ONE7, 1–10 (2012)

    2012

  11. [19]

    Proulx, R.et al.Diversity promotes temporal stability across levels of ecosystem orga- nization in experimental grasslands.PLOS ONE5, 1–8 (2010)

    2010

  12. [20]

    W., King, A

    Jolles, J. W., King, A. J. & Killen, S. S. The role of individual heterogeneity in collective animal behaviour.Trends in Ecology & Evolution35, 278–291 (2020)

    2020

  13. [21]

    & Be’er, A

    Benisty, S., Ben-Jacob, E., Ariel, G. & Be’er, A. Antibiotic-induced anomalous statistics of collective bacterial swarming.Physical Review Letters114, 018105 (2015)

    2015

  14. [22]

    & Ingham, C

    Finkelshtein, A., Roth, D., Ben Jacob, E. & Ingham, C. J. Bacterial swarms recruit cargo bacteria to pave the way in toxic environments.MBio6, 10–1128 (2015)

    2015

  15. [23]

    Zuo, W. & Wu, Y. Dynamic motility selection drives population segregation in a bac- terial swarm.Proceedings of the National Academy of Sciences117, 4693–4700 (2020)

    2020

  16. [24]

    J., Pitchford, J

    Bottery, M. J., Pitchford, J. W. & Friman, V.-P. Ecology and evolution of antimicrobial resistance in bacterial communities.The ISME journal15, 939–948 (2021)

    2021

  17. [25]

    Peled, S.et al.Heterogeneous bacterial swarms with mixed lengths.Physical Review E 103, 032413 (2021)

    2021

  18. [26]

    & Weidenm¨ uller, A

    Jeanson, R. & Weidenm¨ uller, A. Interindividual variability in social insects–proximate causes and ultimate consequences.Biological Reviews89, 671–687 (2014). 18

    2014

  19. [27]

    A., Hunt, E

    O’Shea-Wheller, T. A., Hunt, E. R. & Sasaki, T. Functional heterogeneity in superor- ganisms: Emerging trends and concepts.Annals of the Entomological Society of America 114, 562–574 (2021)

    2021

  20. [28]

    Hemelrijk, C. K. & Kunz, H. Density distribution and size sorting in fish schools: an individual-based model.Behavioral Ecology16, 178–187 (2005)

    2005

  21. [29]

    R., Croft, D

    Dyer, J. R., Croft, D. P., Morrell, L. J. & Krause, J. Shoal composition determines foraging success in the guppy.Behavioral Ecology20, 165–171 (2009)

    2009

  22. [30]

    & Biro, D

    Pettit, B., Akos, Z., Vicsek, T. & Biro, D. Speed determines leadership and leadership determines learning during pigeon flocking.Current Biology25, 3132–3137 (2015)

    2015

  23. [31]

    W., Boogert, N

    Jolles, J. W., Boogert, N. J., Sridhar, V. H., Couzin, I. D. & Manica, A. Consistent individual differences drive collective behavior and group functioning of schooling fish. Current Biology27, 2862–2868 (2017)

    2017

  24. [32]

    del Mar Delgado, M.et al.The importance of individual variation in the dynamics of animal collective movements.Philosophical Transactions of the Royal Society B373, 20170008 (2018)

    2018

  25. [33]

    & Page, S

    Hong, L. & Page, S. E. Groups of diverse problem solvers can outperform groups of high-ability problem solvers.Proceedings of the National Academy of Sciences101, 16385–16389 (2004)

    2004

  26. [34]

    J., Ruxton, G

    Krause, S., James, R., Faria, J. J., Ruxton, G. D. & Krause, J. Swarm intelligence in humans: Diversity can trump ability.Animal Behaviour81, 941–948 (2011)

    2011

  27. [35]

    & Surucu, O

    LiCalzi, M. & Surucu, O. The power of diversity over large solution spaces.Management Science58, 1408–1421 (2012)

    2012

  28. [36]

    S., Xu, H., Jiang, A

    Marcolino, L. S., Xu, H., Jiang, A. X., Tambe, M. & Bowring, E. Give a hard problem to a diverse team: Exploring large action spaces. InProceedings of the AAAI Conference on Artificial Intelligence(2014). 19

    2014

  29. [37]

    Aminpour, P.et al.The diversity bonus in pooling local knowledge about complex problems.Proceedings of the National Academy of Sciences118, e2016887118 (2021)

    2021

  30. [38]

    Physical Review Letters125, 218302 (2020)

    Karamched, B.et al.Heterogeneity improves speed and accuracy in social networks. Physical Review Letters125, 218302 (2020)

    2020

  31. [39]

    & Navajas, J

    Barrera-Lemarchand, F., Balenzuela, P., Bahrami, B., Deroy, O. & Navajas, J. Promot- ing erroneous divergent opinions increases the wisdom of crowds.Psychological Science 35, 872–886 (2024)

    2024

  32. [40]

    When are two heads better than one and why?Science336, 360–362 (2012)

    Koriat, A. When are two heads better than one and why?Science336, 360–362 (2012)

    2012

  33. [41]

    & Ganser, C

    Keuschnigg, M. & Ganser, C. Crowd wisdom relies on agents’ ability in small groups with a voting aggregation rule.Management Science63, 818–828 (2017)

    2017

  34. [42]

    Choudhary, V., Marchetti, A., Shrestha, Y. R. & Puranam, P. Human-AI ensembles: When can they work?Journal of Management51, 536–569 (2025)

    2025

  35. [43]

    Giraldeau, L.-A., Valone, T. J. & Templeton, J. J. Potential disadvantages of using so- cially acquired information.Philosophical Transactions of the Royal Society of London. Series B357, 1559–1566 (2002)

    2002

  36. [44]

    & Ferrante, E

    Prasetyo, J., De Masi, G., Ranjan, P. & Ferrante, E. The best-of-n problem with dynamic site qualities: Achieving adaptability with stubborn individuals. InProceedings of the International Conference on Swarm Intelligence, 239–251 (Springer, 2018)

    2018

  37. [45]

    & Laland, K

    Toyokawa, W., Whalen, A. & Laland, K. N. Social learning strategies regulate the wisdom and madness of interactive crowds.Nature Human Behaviour3, 183–193 (2019)

    2019

  38. [46]

    & Gilmour, D

    Lecaudey, V. & Gilmour, D. Organizing moving groups during morphogenesis.Current Opinion in Cell Biology18, 102–107 (2006)

    2006

  39. [47]

    & Cohen, D

    Shim, G., Devenport, D. & Cohen, D. J. Overriding native cell coordination enhances external programming of collective cell migration.Proceedings of the National Academy of Sciences118, e2101352118 (2021). 20

    2021

  40. [48]

    Rutherford, S. T. & Bassler, B. L. Bacterial quorum sensing: Its role in virulence and possibilities for its control.Cold Spring Harbor perspectives in medicine2, a012427 (2012)

    2012

  41. [49]

    & Reggiani, C

    Bottinelli, R. & Reggiani, C. Human skeletal muscle fibres: Molecular and functional diversity.Progress in Biophysics and Molecular Biology73, 195–262 (2000)

    2000

  42. [50]

    McLean, D. L. & Dougherty, K. J. Peeling back the layers of locomotor control in the spinal cord.Current Opinion in Neurobiology33, 63–70 (2015)

    2015

  43. [51]

    Purslow, P. P. The structure and role of intramuscular connective tissue in muscle function.Frontiers in Physiology11, 495 (2020)

    2020

  44. [52]

    & Farina, D

    Hug, F., Avrillon, S., Ib´ a˜ nez, J. & Farina, D. Common synaptic input, synergies and size principle: Control of spinal motor neurons for movement generation.The Journal of Physiology601, 11–20 (2023)

    2023

  45. [53]

    How physical interactions shape bacterial biofilms.Annual Review of Bio- physics50, 401–417 (2021)

    Maier, B. How physical interactions shape bacterial biofilms.Annual Review of Bio- physics50, 401–417 (2021)

    2021

  46. [54]

    D., Krause, J., Franks, N

    Couzin, I. D., Krause, J., Franks, N. R. & Levin, S. A. Effective leadership and decision- making in animal groups on the move.Nature433, 513–516 (2005)

    2005

  47. [55]

    Gelblum, A.et al.Ant groups optimally amplify the effect of transiently informed individuals.Nature Communications6, 7729 (2015)

    2015

  48. [56]

    Fonio, E.et al.A locally-blazed ant trail achieves efficient collective navigation despite limited information.Elife5, e20185 (2016)

    2016

  49. [57]

    M.et al.Collective animal navigation and migratory culture: From theo- retical models to empirical evidence.Philosophical Transactions of the Royal Society B 373, 20170009 (2018)

    Berdahl, A. M.et al.Collective animal navigation and migratory culture: From theo- retical models to empirical evidence.Philosophical Transactions of the Royal Society B 373, 20170009 (2018)

    2018

  50. [58]

    Kao, A. B. & Couzin, I. D. Modular structure within groups causes information loss but can improve decision accuracy.Philosophical Transactions of the Royal Society B 374, 20180378 (2019). 21

    2019

  51. [59]

    Almaatouq, A.et al.Adaptive social networks promote the wisdom of crowds.Proceed- ings of the National Academy of Sciences117, 11379–11386 (2020)

    2020

  52. [60]

    Galesic, M.et al.Beyond collective intelligence: Collective adaptation.Journal of the Royal Society Interface20, 20220736 (2023)

    2023

  53. [61]

    Glassman, R. B. Persistence and loose coupling in living systems.Behavioral Science 18, 83–98 (1973)

    1973

  54. [62]

    & Salamon, P

    Hansen, L. & Salamon, P. Neural network ensembles.IEEE Transactions on Pattern Analysis and Machine Intelligence12, 993–1001 (1990)

    1990

  55. [63]

    Dietterich, T. G. Ensemble methods in machine learning. InInternational Workshop on Multiple Classifier Systems, 1–15 (Springer, 2000)

    2000

  56. [64]

    Dorigo, M.et al.Swarmanoid: A novel concept for the study of heterogeneous robotic swarms.IEEE Robotics & Automation Magazine20, 60–71 (2013)

    2013

  57. [65]

    & Hamann, H

    Raoufi, M., Romanczuk, P. & Hamann, H. Individuality in swarm robots with the case study of kilobots: Noise, bug, or feature? InProceedings of the Artificial Life Conference (ALIFE), 1, 35 (MIT Press, 2023)

    2023

  58. [66]

    & Kiviranta, M

    M¨ ammel¨ a, A., Riekki, J. & Kiviranta, M. Loose coupling: An invisible thread in the history of technology.IEEE Access11, 59456–59482 (2023)

    2023

  59. [67]

    & Ebrahimpour, R

    Masoudnia, S. & Ebrahimpour, R. Mixture of experts: A literature survey.Artificial Intelligence Review42, 275–293 (2014)

    2014

  60. [68]

    InProceedings of the International Conference on Learning Representa- tions, (ICLR)(2017)

    Shazeer, N.et al.Outrageously large neural networks: The sparsely-gated Mixture-of- Experts layer. InProceedings of the International Conference on Learning Representa- tions, (ICLR)(2017)

    2017

  61. [69]

    & Aracil, R

    Baca, J., Ferre, M. & Aracil, R. A heterogeneous modular robotic design for fast response to a diversity of tasks.Robotics and Autonomous Systems60, 522–531 (2012). 22

    2012

  62. [70]

    S., Saha, A., Marshall, J

    Talamali, M. S., Saha, A., Marshall, J. A. & Reina, A. When less is more: Robot swarms adapt better to changes with constrained communication.Science Robotics6, eabf1416 (2021)

    2021

  63. [71]

    S., Dorigo, M., Hamann, H

    Aust, T., Talamali, M. S., Dorigo, M., Hamann, H. & Reina, A. The hidden ben- efits of limited communication and slow sensing in collective monitoring of dynamic environments. InInternational Conference on Swarm Intelligence, 234–247 (Springer, 2022)

    2022

  64. [72]

    & Reina, A

    Zakir, R., Salahshour, M., Dorigo, M. & Reina, A. Heterogeneity can enhance the adaptivity of robot swarms to dynamic environments. InProceedings of the International Conference on Swarm Intelligence, 112–126 (Springer, 2024)

    2024

  65. [73]

    J., Hussein, A

    Hepworth, A. J., Hussein, A. S., Reid, D. J. & Abbass, H. A. Contextually aware intelligent control agents for heterogeneous swarms.Swarm Intelligence18, 275–310 (2024)

    2024

  66. [74]

    & Trianni, V

    Dorigo, M., Theraulaz, G. & Trianni, V. Swarm robotics: Past, present, and future [point of view].Proceedings of the IEEE109, 1152–1165 (2021)

    2021

  67. [75]

    & Bender, A

    Whitacre, J. & Bender, A. Pervasive flexibility in living technologies through degeneracy-based design.Artificial Life19, 365–386 (2013)

    2013

  68. [76]

    & Elmenreich, W

    Schranz, M., Umlauft, M., Sende, M. & Elmenreich, W. Swarm robotic behaviors and current applications.Frontiers in Robotics and AI7, 36 (2020)

    2020

  69. [77]

    Lehman, J., Meyerson, E., El-Gaaly, T., Stanley, K. O. & Ziyaee, T. Evolution and the Knightian blindspot of machine learning.CoRRabs/2501.13075(2025)

    2025 arXiv

  70. [78]

    & Wyffels, F

    Degrave, J., Hermans, M., Dambre, J. & Wyffels, F. A differentiable physics engine for deep learning in robotics.Frontiers in Neurorobotics13, 6 (2019)

    2019

  71. [79]

    InPro- ceedings of the International Conference on Learning Representations (ICLR)(2020)

    Hu, Y.et al.Difftaichi: Differentiable programming for physical simulation. InPro- ceedings of the International Conference on Learning Representations (ICLR)(2020). 23

    2020

  72. [80]

    Evolving 3D morphology and behavior by competition.Artificial Life1, 353–372 (1994)

    Sims, K. Evolving 3D morphology and behavior by competition.Artificial Life1, 353–372 (1994)

    1994

  73. [81]

    & Pollack, J

    Lipson, H. & Pollack, J. B. Automatic design and manufacture of robotic lifeforms. Nature406, 974–978 (2000)

    2000

  74. [82]

    & Nolfi, S

    Floreano, D., Husbands, P. & Nolfi, S. Evolutionary robotics. In Siciliano, B. & Khatib, O. (eds.)Springer Handbook of Robotics, 1423–1451 (Springer, Berlin, Germany, 2008)

    2008

  75. [83]

    & Stanley, K

    Lehman, J. & Stanley, K. O. Evolving a diversity of virtual creatures through novelty search and local competition. InProceedings of the Genetic and Evolutionary Compu- tation Conference (GECCO), 211–218 (2011)

    2011

  76. [84]

    & Bongard, J

    Matthews, D., Spielberg, A., Rus, D., Kriegman, S. & Bongard, J. Efficient automatic design of robots.Proceedings of the National Academy of Sciences120, e2305180120 (2023)

    2023

  77. [85]

    & Shaposhnik, Y

    Wang, Y., Huang, H., Rudin, C. & Shaposhnik, Y. Understanding how dimension reduction tools work: an empirical approach to deciphering t-SNE, UMAP, TriMAP, and PaCMAP for data visualization.Journal of Machine Learning Research22, 1–73 (2021)

    2021

  78. [86]

    Li, S.et al.Particle robotics based on statistical mechanics of loosely coupled compo- nents.Nature567, 361–365 (2019)

    2019

  79. [87]

    body genotype

    Savoie, W.et al.A robot made of robots: Emergent transport and control of a smarticle ensemble.Science robotics4, eaax4316 (2019). 24 7 Supplementary Information 7.1 Supplementary Methods 7.1.1 Agents. Each agent consists of 11 point masses (n P M) connected by 21 springs in a...

    2019

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.