Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Emergent Heterogeneous Swarm Control Through Hebbian Learning

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that evolving a single set of Hebbian learning rules—local weight-update equations shared identically by every robot—can replace both homogeneous controllers and multi-agent reinforcement learning for heterogeneous…

desk verdict First real application of evolved Hebbian rules to heterogeneous swarm control, with a solid MARL comparison and hardware transfer, but the headline 'emergent heterogeneity' claim is confounded with per-agent random initialization and needs one control experiment. read the letter →

arxiv 2507.11566 v1 pith:DQOPXPU6 submitted 2025-07-14 cs.NE cs.AIcs.RO

Fuda van Diggelen , Tugay Alperen Karagüzel , Andres Garcia Rincon , A.E. Eiben , Dario Floreano , Eliseo Ferrante This is my paper · ORCID
classification cs.NEcs.AIcs.RO
keywords SwarmRoboticsHebbianLearningEmergenceHeterogeneousswarmsNeuroevolutionMulti-AgentReinforcementSourcelocalisationSim-to-realtransfer
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 claims that evolving a single set of Hebbian learning rules—local weight-update equations shared identically by every robot—can replace both hand-designed homogeneous controllers and multi-agent reinforcement learning for heterogeneous swarms. Because each agent's network starts with independent random weights and updates them using only its own local sensory activity, individual controllers diverge automatically, so specialisation emerges without predefined roles or a centralised trainer. The authors report that this method significantly outperforms homogeneous evolution and two MARL algorithms (MADDPG and MATD3) on the PettingZoo benchmarks multiwalker v9 and waterworld v4, and outperforms the baselines on a simulated and real-robot source-localisation task. They also observe emergent swarm-level behavioural switching: the swarm first circles to sense the light gradient, then moves in a straight line toward the source, and re-adapts after the source is moved. A sympathetic reader would take the paper's central claim to be that local correlation-based learning rules are a scalable, parameter-efficient route to heterogeneous swarm control.

What carries the argument

The load-bearing mechanism is the Hebbian ABCD update rule of Equation 1, a local correlation rule $\delta w_{i,j} = \mu(A_{i,j}n_i n_j + B_{i,j}n_i + C_{i,j}n_j + D_{i,j})$ with fixed learning rate $\mu=0.1$ and per-weight coefficients $A,B,C,D \in [-5,5]$ that are evolved by CMA-ES. Copies of the same coefficient set are given to every agent, so the optimisation cost is $4\times|W|$ parameters (720 in the swarm experiments) and is independent of swarm size; the rule converts each agent's private sensory history into private weight changes, which is what turns a uniform rule into heterogeneous controllers.

What would settle it

A decisive test is to rerun the source-localisation experiment with the ABCD rules disabled ($\delta w = 0$ for all weights) while keeping independent random initial weights and all other settings identical. If the swarm's final fitness and between-agent weight variance remain comparable to the Hebbian condition, the claim that the learning rules drive emergent heterogeneity would be falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that heterogeneity in a robot swarm can be a by-product of a single evolved Hebbian update rule rather than a designed property. Each robot runs the same network architecture and the same ABCD rule $\delta w_{i,j} = \mu(A_{i,j}n_i n_j + B_{i,j}n_i + C_{i,j}n_j + D_{i,j})$, but because the initial weights are sampled independently and the rule acts only on locally sensed pre- and post-synaptic activity, each robot's weights follow a unique trajectory. The resulting controllers differ across the swarm, swarm-level behaviour switches between gradient sensing and goal approach, and fitness is higher than homogeneous evolution and MARL baselines. In the real-world transfer, the Hebbian controller kept roughly the same average fitness while the baselines dropped by more than ten percent.

Load-bearing premise

The argument assumes the performance and diversity of the Hebbian swarm come from the evolved learning rules, but the Hebbian agents also start with independently random network weights, so the observed heterogeneity could in principle be caused by that random initialisation alone.

Editorial extensions

If this is right

  • Evolving update rules rather than controllers makes the optimisation cost independent of swarm size, so the method scales to larger swarms where MARL's joint action-state space would explode, as demonstrated in waterworld v4.
  • Heterogeneous roles need no longer be specified in advance: the same rule can generate different specialisations through local experience, and can re-configure when the task or environment changes.
  • Hebbian controllers retain the ability to switch collective behaviours online, including recovering after an abrupt environmental perturbation, without any retraining.
  • The small measured sim-to-real gap (-0.7% for Hebbian versus more than 10% drops for baselines) suggests evolved learning rules are a practical route to real-robot deployment.
  • The method matches or beats MADDPG and MATD3 on the tested PettingZoo benchmarks, positioning evolved Hebbian learning as a viable MARL alternative in continuous-control swarm tasks.

Reading between the lines

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

  • Because the method needs no communication and only local sensing, the same evolved-rule scheme should transfer to other collective tasks (for example collective transport or exploration) whenever a swarm-level fitness can be defined; this is an extension the paper does not test.
  • The observed weight trajectories that keep diverging without converging suggest the network itself acts as a slow memory channel; exploiting that channel could support continual learning and fast re-adaptation without experience replay.
  • Adding weight normalisation (for instance an Oja-style term) to the ABCD rule would likely prevent the unbounded weight growth the paper reports, and could be tested as a drop-in modification within the same framework.
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

4 major / 5 minor

Summary. The paper proposes evolving a single set of Hebbian ABCD update rules (Eq. 1) shared by all swarm members; each agent starts with an independently random neural network and updates its weights online from local sensor correlations. The central claims are that heterogeneity emerges automatically from these uniform rules, that the approach outperforms homogeneous evolution and two MARL baselines (MADDPG, MATD3) on Pettingzoo benchmarks, and that the evolved controller transfers to a real Thymio swarm with a small sim-to-real gap. Additional analyses address scalability, flexibility, behavioural switching, and weight dynamics.

Significance. If the central causal claim is established, this is a valuable contribution: a fixed number of evolved parameters independent of swarm size, a local and decentralised learning mechanism, and an apparent alternative to MARL in standard benchmarks. The paper has concrete strengths: code and data links are provided, the MARL comparisons use 30 repetitions, the evolution experiments use 10 runs, validation experiments use N=60, real-robot transfer is attempted, and the Discussion is candid about the possibility that heterogeneity is not guaranteed and that random initialisation may be exploited early in adaptation. However, the paper's headline claim that Hebbian learning causes heterogeneity to emerge is not yet supported because the Hebbian condition is confounded with per-agent random initialisation; this issue is load-bearing and needs additional controls.

major comments (4)
  1. [§4.4 and §4.5] The comparison between the Hebbian condition and the Baseline is confounded. Section 4.4 states that 'each swarm member receives the same ABCD rules and initialises the NN controller randomly U[-1,1]', while Section 4.5 defines the Baseline as assigning 'the same NN controller' to every member with weights evolved between trials. Thus the Hebbian condition changes two things at once: per-agent random initialisation and online Hebbian updates. The heterogeneity metric in Section 4.6.3 (Eq. 5) starts at a nonzero value because of this random initialisation and then grows, so it cannot by itself attribute divergence to the Hebbian rules. The Discussion's admission that 'Hebbian learning might transiently exploit random initialisation to locate gradients during early adaptation' makes this concern concrete. I request at least two controls: (i) shared initial weights with Hebbian updates, and (ii) per-agent random initial weights with no online updates; ideally also (iii) per-agent random initial weights with a non-Hebbian local update rule. Without these, the abstract claim that Hebbian learning 'enables the automatic emergence of heterogeneity' is not established.
  2. [Tables A2 and A3] There is an internal inconsistency in the reported performance of the Hebbian-1 controller in the original circular environment. Table A2 reports Hebbian-1 as 0.22 ± 0.031 (N=60) for the 'original' column, while Table A3 reports Hebbian-1 as 0.83 ± 0.040 (N=60) for the same 'original' circular condition. Both tables refer to the same best controller and the same environment, so at least one entry is incorrect. Since the aggregate significance claims in Appendix A combine results from these tables, the error is load-bearing for the additional controller comparisons and must be corrected before the related claims can be evaluated.
  3. [§2.2 and Figure 6b] The real-world reality-gap claim is based on only three runs per controller, with no error bars and no statistical test. The text states that the Hebbian approach 'maintains similar average fitness (-0.7%)' while both Baseline methods drop 'more than 10 percent', but Figure 6b shows no variability for the three real-world repetitions and the comparison to the simulation results (N=60) is informal. This claim should either be supported by more real-world repetitions and a significance test, or it should be explicitly framed as a qualitative, preliminary observation rather than a quantitative result.
  4. [§2.1 and Table 1] The statistical reporting for the MARL benchmarks is incomplete. The text reports p < 0.001 and p < 0.005 without naming the test, whether it was two-tailed, or how multiple comparisons were handled; for example, the waterworld p-value is reported as p < 0.005, which is unusual for a single comparison. The numerical formatting also appears corrupted ('Baseline−275.56.8±32.6'). Please specify the test and corrections, and provide effect sizes or the underlying distributions, because the central performance comparison rests on these numbers.
minor comments (5)
  1. [Throughout] There are several typos and formatting errors, including 'environmenst' in Section 2.1, 'decribed' in Section 2.1, 'simualted' in Section 2.2, and 'repititions' in Figure 6b; these should be corrected.
  2. [§2.1] The sentence 'For MARL algorithms we found similar results as decribed in .' contains an empty reference placeholder and should be completed or removed.
  3. [§5] The Data Availability statement says the data 'will be deposited' at a future URL; for a published claim of reproducibility, the data should be available at the time of publication or the statement should state a clear embargo period.
  4. [Table 1] The text in Section 4.1 says the NN size is chosen so that the number of optimisation parameters for the actor is approximately equal to the Hebbian condition, but Table 1 lists Baseline parameters as 560/3904 and Hebbian parameters as 2240/15616 for the two environments; the discrepancy between the text and the table should be clarified.
  5. [§3] The statement that 'Hebbian learning consistently outperforms across different tasks' is supported for the tested tasks, but the negative correlation with scalability mentioned in the Discussion is not quantified; adding the correlation coefficient and test would make the observation more precise.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the main performance claims are benchmarked against external MARL algorithms and a homogeneous baseline; the only substantive concern is an acknowledged random-initialisation confound in the heterogeneity-emergence claim, plus minor non-load-bearing self-citations.

full rationale

The derivation chain is not circular. The central performance claim compares Hebbian learning against MADDPG, MATD3, and a homogeneous Baseline in PettingZoo environments, and against a homogeneous Baseline and Baseline-A in the source-localisation task; these benchmarks are external or at least not derived from the paper's own fitted outputs. The evolved ABCD rules are optimised directly on swarm-level fitness, and the observed performance differences are empirical results, not identities with the training objective. The only notable weakness is an experimental confound, not a circular step: Section 4.4 states that 'each swarm member receives the same ABCD rules and initialises the NN controller randomly U[-1,1]', while Section 4.5 defines the Baseline as assigning 'the same NN controller' to every member. Consequently the heterogeneity metric in Equation 5 is already nonzero at t=0, and the Discussion explicitly concedes that 'Hebbian learning might transiently exploit random initialisation to locate gradients during early adaptation.' This means the causal claim that Hebbian updates alone produce the observed heterogeneity is not fully isolated; a shared-initialisation Hebbian condition or a random-initialisation Baseline control would be needed. However, this is a confound rather than a circular reduction: the reported increase in weight STD is an empirical observation that the paper itself notes is not guaranteed, and the main performance gains are externally validated. The self-citations present ([48], [52], [46]) are used as a comparison baseline, prior behavioural context, and reality-gap reference respectively; none is load-bearing for the central claims. Overall, no prediction is equivalent to its own input by construction, so circularity is minor and localised to the interpretation of the heterogeneity result.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The method's performance is produced by fitting a large number of free parameters (the ABCD rules) to the task via CMA-ES. The learning rule form, fitness function, noise model, and hyperparameters are assumed. No new physical entities are introduced.

free parameters (6)
  • ABCD Hebbian rules = 720 for source localisation; 2240 (multiwalker), 15616 (waterworld)
    Evolved via CMA-ES to maximize swarm-level fitness; this is the core fitted content of the method.
  • Learning rate mu = 0.1
    Fixed constant, not evolved; chosen by hand.
  • ABCD rule bounds = [-5,5] (mutation range; initial U[-0.1,0.1])
    Chosen bounds for the evolved learning rules; not derived.
  • CMA-ES hyperparameters = sigma0=1.0, population 30, generations 100
    Standard optimiser settings, chosen, not derived.
  • Baseline-A state-machine thresholds = light thresholds 229/178/127 and probabilities 1.0/0.75/0.5/0.25
    Obtained by retesting the controller across swarm ratios and distances (Table 3); fitted to the task.
  • Sensor noise model = light=0.05, theta=0.043, d=0.0046, 20% faulty readings
    Fit to real-world measurements for the simulator; a modelling choice.
assumptions (4)
  • ad hoc to paper The linear-correlation ABCD update rule (Eq. 1) is the mechanism for all learning.
    The paper postulates this specific form of Hebbian plasticity rather than deriving it from first principles.
  • domain assumption Swarm-level fitness (average light intensity, Eq. 2) is the right objective for source localisation.
    The choice of average light intensity as fitness is a task-specific proxy, not derived.
  • domain assumption CMA-ES optimisation of the learning rules converges to a meaningful optimum within 100 generations.
    Standard evolutionary robotics practice; no convergence guarantee is given.
  • domain assumption The simulator noise model matches the real Thymio swarm closely enough for sim-to-real transfer.
    The sensor noise parameters are fit to real measurements, but the validity of this fit is not quantified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Emergent Heterogeneous Swarm Control Through Hebbian Learning." pith.science (2026). https://pith.science/paper/DQOPXPU6

@misc{pith2026250711566,
  author       = {Pith},
  title        = {Pith review of: Emergent Heterogeneous Swarm Control Through Hebbian Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DQOPXPU6}},
  note         = {Machine review of arXiv:2507.11566}
}
read the original abstract

In this paper, we introduce Hebbian learning as a novel method for swarm robotics, enabling the automatic emergence of heterogeneity. Hebbian learning presents a biologically inspired form of neural adaptation that solely relies on local information. By doing so, we resolve several major challenges for learning heterogeneous control: 1) Hebbian learning removes the complexity of attributing emergent phenomena to single agents through local learning rules, thus circumventing the micro-macro problem; 2) uniform Hebbian learning rules across all swarm members limit the number of parameters needed, mitigating the curse of dimensionality with scaling swarm sizes; and 3) evolving Hebbian learning rules based on swarm-level behaviour minimises the need for extensive prior knowledge typically required for optimising heterogeneous swarms. This work demonstrates that with Hebbian learning heterogeneity naturally emerges, resulting in swarm-level behavioural switching and in significantly improved swarm capabilities. It also demonstrates how the evolution of Hebbian learning rules can be a valid alternative to Multi Agent Reinforcement Learning in standard benchmarking tasks.

Figures

Figures reproduced from arXiv: 2507.11566 by the authors.

Figure 2
Figure 2. [PITH_FULL_IMAGE:figures/full_fig_p028_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Research Novelty in Information Systems Journals After ChatGPT: Differences Across Institutional Language Contexts

    cs.DL 2026-03 unverdicted novelty 6.0 of 10

    Post-2022, IS articles from non-English-dominant first-author institutions show a 0.176 SD larger decline in relative semantic novelty (~7 percentile points) than English-dominant ones.

Reference graph

Works this paper leans on

57 extracted references · 48 canonical work pages · cited by 1 Pith paper

  1. [1]

    SIGGRAPH Comput

    Reynolds, C.W.: Flocks, herds and schools: A distributed behav- ioral model. SIGGRAPH Comput. Graph. 21(4), 25–34 (1987) https://doi.org/10.1145/37402.37406

    arXiv 1987

  2. [2]

    Nature 519(7542), 233–236 (2015)

    Ramdya, P., Lichocki, P., Cruchet, S., Frisch, L., Tse, W., Floreano, D., Benton, R.: Mechanosensory interactions drive collective behaviour in drosophila. Nature 519(7542), 233–236 (2015)

    work page 2015

  3. [3]

    Scientific reports 8(1), 1–11 (2018)

    Puckett, J.G., Pokhrel, A.R., Giannini, J.A.: Collective gradient sensing in fish schools. Scientific reports 8(1), 1–11 (2018)

    work page 2018

  4. [4]

    Nature Communications 13(1), 2315 (2022)

    Cavagna, A., Culla, A., Feng, X., Giardina, I., Grigera, T.S., Kion-Crosby, W., Melillo, S., Pisegna, G., Postiglione, L., Villegas, P.: Marginal speed confine- ment resolves the conflict between correlation and control in collective behaviour. Nature Communications 13(1), 2315 (2022)

    work page 2022

  5. [5]

    Nature communications 12(1), 4345 (2021)

    Hasselmann, K., Ligot, A., Ruddick, J., Birattari, M.: Empirical assessment and comparison of neuro-evolutionary methods for the automatic off-line design of robot swarms. Nature communications 12(1), 4345 (2021)

    work page 2021

  6. [6]

    Frontiers in Robotics and AI 12, 1607978 (2025)

    Kegeleirs, M., Birattari, M.: Towards applied swarm robotics: current limitations and enablers. Frontiers in Robotics and AI 12, 1607978 (2025)

    work page 2025

  7. [7]

    Proceedings of the IEEE 109(7), 1152–1165 (2021)

    Dorigo, M., Theraulaz, G., Trianni, V.: Swarm robotics: Past, present, and future [point of view]. Proceedings of the IEEE 109(7), 1152–1165 (2021)

    work page 2021

  8. [8]

    Swarm Intelligence 7, 1–41 (2013)

    Brambilla, M., Ferrante, E., Birattari, M., Dorigo, M.: Swarm robotics: A review from the swarm engineering perspective. Swarm Intelligence 7, 1–41 (2013)

    work page 2013

Show all 57 references
  1. [9]

    Neurocomputing 172, 292–321 (2016)

    Bayındır, L.: A review of swarm robotics tasks. Neurocomputing 172, 292–321 (2016)

    2016

  2. [10]

    (eds.) AutoMoDe: A Modular Approach to the Automatic Off-Line Design and Fine-Tuning of Control 20 Software for Robot Swarms, pp

    Birattari, M., Ligot, A., Francesca, G.: In: Pillay, N., Qu, R. (eds.) AutoMoDe: A Modular Approach to the Automatic Off-Line Design and Fine-Tuning of Control 20 Software for Robot Swarms, pp. 73–90. Springer, Cham (2021)

    2021

  3. [11]

    PLoS computational biology 11(8), 1004273 (2015)

    Ferrante, E., Turgut, A.E., Du´ e˜ nez-Guzm´ an, E., Dorigo, M., Wenseleers, T.: Evo- lution of self-organized task specialization in robot swarms. PLoS computational biology 11(8), 1004273 (2015)

    2015

  4. [12]

    In: 2024 IEEE International Conference on Robotics and Automation (ICRA), pp

    Ramos, D.G., Birattari, M.: Automatically designing robot swarms in envi- ronments populated by other robots: an experiment in robot shepherding. In: 2024 IEEE International Conference on Robotics and Automation (ICRA), pp. 12240–12247 (2024). https://doi.org/10.1109/ICRA57147...

    2024

  5. [13]

    ALIFE 2023: Ghost in the Machine: Proceedings of the 2023 Artificial Life Conference, p

    Raoufi, M., Romanczuk, P., Hamann, H.: Individuality in Swarm Robots with the Case Study of Kilobots: Noise, Bug, or Feature? ALIFE 2022: The 2022 Con- ference on Artificial Life, vol. ALIFE 2023: Ghost in the Machine: Proceedings of the 2023 Artificial Life Conference, p. 35 ...

    2023

  6. [14]

    Scientific Reports 14(1), 27314 (2024)

    Mengers, V., Raoufi, M., Brock, O., Hamann, H., Romanczuk, P.: Leveraging uncertainty in collective opinion dynamics with heterogeneity. Scientific Reports 14(1), 27314 (2024)

    2024

  7. [15]

    The horizons of evolutionary robotics 153 (2014)

    Trianni, V., Tuci, E., Ampatzis, C., Dorigo, M.: Evolutionary swarm robotics: A theoretical and methodological itinerary from individual neuro-controllers to collective behaviours. The horizons of evolutionary robotics 153 (2014)

    2014

  8. [16]

    Nature communications 9(1), 931 (2018)

    Scholz, C., Engel, M., P¨ oschel, T.: Rotating robots move collectively and self- organize. Nature communications 9(1), 931 (2018)

    2018

  9. [17]

    Nature Communications 14(1), 3476 (2023)

    Sun, G., Zhou, R., Ma, Z., Li, Y., Groß, R., Chen, Z., Zhao, S.: Mean-shift exploration in shape assembly of robot swarms. Nature Communications 14(1), 3476 (2023)

    2023

  10. [18]

    Frontiers in Robotics and AI 10, 1134841 (2023)

    Kuckling, J.: Recent trends in robot learning and evolution for swarm robotics. Frontiers in Robotics and AI 10, 1134841 (2023)

    2023

  11. [19]

    Scientific Reports 14(1), 32019 (2024)

    Li, X., Wang, C., Li, C.: A distributed control strategy for groups of robots with application in flocking. Scientific Reports 14(1), 32019 (2024)

    2024

  12. [20]

    In: Dorigo, M., St¨ utzle, T., Blesa, M.J., Blum, C., Hamann, H., Heinrich, M.K., Strobel, V

    Karag¨ uzel, T.A., Turgut, A.E., Ferrante, E.: Collective gradient perception in a flocking robot swarm. In: Dorigo, M., St¨ utzle, T., Blesa, M.J., Blum, C., Hamann, H., Heinrich, M.K., Strobel, V. (eds.) Swarm Intelligence, pp. 290–297. Springer, Cham (2020)

    2020

  13. [21]

    In: Hamann, H., Dorigo, M., P´ erez C´ aceres, L., Reina, A., Kuckling, J., Kaiser, T.K., Soorati, M., Hasselmann, K., Buss, E

    Rincon, A.G., Karag¨ uzel, T.A., Diggelen, F.v., Ferrante, E.: Collective random walks of flocking agents through emergent implicit leadership. In: Hamann, H., Dorigo, M., P´ erez C´ aceres, L., Reina, A., Kuckling, J., Kaiser, T.K., Soorati, M., Hasselmann, K., Buss, E. (eds....

    2024

  14. [22]

    Springer, Cham (2016)

    Nolfi, S., Bongard, J., Husbands, P., Floreano, D.: Evolutionary Robotics. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-32552-1 76

    2016 doi

  15. [23]

    Springer-Verlag, London (2015)

    Eiben, A.E., Smith, J.E.: Introduction to Evolutionary Computing, 2nd edn. Springer-Verlag, London (2015)

    2015

  16. [24]

    Nature communications 15(1), 6267 (2024)

    Diggelen, F., Cambier, N., Ferrante, E., Eiben, A.: A model-free method to learn multiple skills in parallel on modular robots. Nature communications 15(1), 6267 (2024)

    2024

  17. [25]

    Plos one 14(10), 0224376 (2019)

    Ramos, R.P., Oliveira, S.M., Vieira, S.M., Christensen, A.L.: Evolving flocking in embodied agents based on local and global application of reynolds’ rules. Plos one 14(10), 0224376 (2019)

    2019

  18. [26]

    IEEE Robotics and Automation Letters 4(4), 4523–4530 (2019)

    Schilling, F., Lecoeur, J., Schiano, F., Floreano, D.: Learning vision-based flight in drone swarms by imitation. IEEE Robotics and Automation Letters 4(4), 4523–4530 (2019)

    2019

  19. [27]

    Sensors 23(7), 3625 (2023)

    Orr, J., Dutta, A.: Multi-agent deep reinforcement learning for multi-robot applications: A survey. Sensors 23(7), 3625 (2023)

    2023

  20. [28]

    IEEE transactions on Evolutionary Computation 13(3), 648–660 (2009)

    Waibel, M., Keller, L., Floreano, D.: Genetic team composition and level of selection in the evolution of cooperation. IEEE transactions on Evolutionary Computation 13(3), 648–660 (2009)

    2009

  21. [29]

    Philosophical Transactions of the Royal Society B 377(1843), 20200309 (2022)

    Bredeche, N., Fontbonne, N.: Social learning in swarm robotics. Philosophical Transactions of the Royal Society B 377(1843), 20200309 (2022)

    2022

  22. [30]

    AAMAS ’23, pp

    Bettini, M., Shankar, A., Prorok, A.: Heterogeneous multi-robot reinforcement learning. AAMAS ’23, pp. 1485–1494. International Foundation for Autonomous Agents and Multiagent Systems, Richland, SC (2023)

    2023

  23. [31]

    MIT Press, Cambridge, MA (2024)

    Albrecht, S.V., Christianos, F., Sch¨ afer, L.: Multi-Agent Reinforcement Learning: Foundations and Modern Approaches. MIT Press, Cambridge, MA (2024)

    2024

  24. [32]

    arXiv preprint arXiv:2203.05285 (2022)

    Hao, X., Mao, H., Wang, W., Yang, Y., Li, D., Zheng, Y., Wang, Z., Hao, J.: Breaking the curse of dimensionality in multiagent state space: A unified agent permutation framework. arXiv preprint arXiv:2203.05285 (2022)

    2022 arXiv

  25. [33]

    Wiley, New York (1949)

    Hebb, D.O.: The Organization of Behavior: A Neuropsychological Theory. Wiley, New York (1949)

    1949

  26. [34]

    Biological cybernetics 87(5), 404–415 (2002)

    Gerstner, W., Kistler, W.M.: Mathematical formulations of hebbian learning. Biological cybernetics 87(5), 404–415 (2002)

    2002

  27. [35]

    Advances in Neural Information Processing Systems 33, 20719–20731 (2020) 22

    Najarro, E., Risi, S.: Meta-learning through hebbian plasticity in random net- works. Advances in Neural Information Processing Systems 33, 20719–20731 (2020) 22

    2020

  28. [36]

    IEEE Transactions on Cognitive and Developmental Systems 15(3), 1536–1546 (2022)

    Ferigo, A., Iacca, G., Medvet, E., Pigozzi, F.: Evolving hebbian learning rules in voxel-based soft robots. IEEE Transactions on Cognitive and Developmental Systems 15(3), 1536–1546 (2022)

    2022

  29. [37]

    Neurocomputing 614, 128811 (2025)

    Ferigo, A., Iacca, G., Medvet, E., Nadizar, G.: Totipotent neural controllers for modular soft robots: Achieving specialization in body–brain co-evolution through hebbian learning. Neurocomputing 614, 128811 (2025)

    2025

  30. [38]

    In: 2007 IEEE Congress on Evolutionary Computation, pp

    Soltoggio, A., Durr, P., Mattiussi, C., Floreano, D.: Evolving neuro- modulatory topologies for reinforcement learning-like problems. In: 2007 IEEE Congress on Evolutionary Computation, pp. 2471–2478 (2007). https://doi.org/10.1109/CEC.2007.4424781 . IEEE

    2007

  31. [39]

    IEEE Transactions on evolutionary computation 11(5), 596–607 (2007)

    Mattiussi, C., Floreano, D.: Analog genetic encoding for the evolution of circuits and networks. IEEE Transactions on evolutionary computation 11(5), 596–607 (2007)

    2007

  32. [40]

    Evolutionary intelligence 1, 47–62 (2008)

    Floreano, D., D¨ urr, P., Mattiussi, C.: Neuroevolution: from architectures to learning. Evolutionary intelligence 1, 47–62 (2008)

    2008

  33. [41]

    : Pettingzoo: Gym for multi-agent reinforcement learning

    Terry, J., Black, B., Grammel, N., Jayakumar, M., Hari, A., Sullivan, R., Santos, L.S., Dieffendahl, C., Horsch, C., Perez-Vicente, R., et al. : Pettingzoo: Gym for multi-agent reinforcement learning. Advances in Neural Information Processing Systems 34, 15032–15043 (2021). Ac...

    2021

  34. [42]

    In: International Conference on Autonomous Agents and Multiagent Systems, pp

    Gupta, J.K., Egorov, M., Kochenderfer, M.: Cooperative multi-agent control using deep reinforcement learning. In: International Conference on Autonomous Agents and Multiagent Systems, pp. 66–83 (2017). Springer

    2017

  35. [43]

    Advances in neural information processing systems 30 (2017)

    Lowe, R., Wu, Y.I., Tamar, A., Harb, J., Pieter Abbeel, O., Mordatch, I.: Multi- agent actor-critic for mixed cooperative-competitive environments. Advances in neural information processing systems 30 (2017)

    2017

  36. [44]

    arXiv preprint arXiv:1910.01465 (2019)

    Ackermann, J., Gabler, V., Osa, T., Sugiyama, M.: Reducing overestimation bias in multi-agent domains using double centralized critics. arXiv preprint arXiv:1910.01465 (2019)

    2019 arXiv

  37. [45]

    In: Proceedings of the Third European Conference on Advances in Artificial Life, pp

    Jacobi, N., Husbands, P., Harvey, I.: Noise and the reality gap: The use of simula- tion in evolutionary robotics. In: Proceedings of the Third European Conference on Advances in Artificial Life, pp. 704–720. Springer, Berlin, Heidelberg (1995)

    1995

  38. [46]

    IEEE Transactions on Cognitive and Developmental Systems 15(2), 499–506 (2021)

    Diggelen, F., Ferrante, E., Harrak, N., Luo, J., Zeeuwe, D., Eiben, A.: The influence of robot traits and evolutionary dynamics on the reality gap. IEEE Transactions on Cognitive and Developmental Systems 15(2), 499–506 (2021)

    2021

  39. [47]

    (eds.) 23 Evolutionary Robotics, pp

    Floreano, D., Husbands, P., Nolfi, S.: In: Siciliano, B., Khatib, O. (eds.) 23 Evolutionary Robotics, pp. 1423–1451. Springer, Berlin, Heidelberg (2008). https://doi.org/10.1007/978-3-540-30301-5 62

    2008 doi

  40. [48]

    In: Affenzeller, M., Winkler, S.M., Kononova, A.V., Trautmann, H., Tuˇ sar, T., Machado, P., B¨ ack, T

    Diggelen, F., Carlo, M., Cambier, N., Ferrante, E., Eiben, G.: Emergence of specialised collective behaviors in evolving heterogeneous swarms. In: Affenzeller, M., Winkler, S.M., Kononova, A.V., Trautmann, H., Tuˇ sar, T., Machado, P., B¨ ack, T. (eds.) Parallel Problem Solvin...

    2024

  41. [49]

    Evolutionary Computation 9(2), 159–195 (2001)

    Hansen, N., Ostermeier, A.: Completely derandomized self-adaptation in evolu- tion strategies. Evolutionary Computation 9(2), 159–195 (2001)

    2001

  42. [50]

    IEEE Robotics & Automation Magazine 24(1), 77–85 (2017)

    Mondada, F., Bonani, M., Riedo, F., Briod, M., Pereyre, L., R´ etornaz, P., Magne- nat, S.: Bringing robotics to formal education: The thymio open-source hardware robot. IEEE Robotics & Automation Magazine 24(1), 77–85 (2017)

    2017

  43. [51]

    Neurocomputing 17(1), 25–45 (1997)

    Oja, E.: The nonlinear pca learning rule in independent component analysis. Neurocomputing 17(1), 25–45 (1997)

    1997

  44. [52]

    In: Proceedings of the Genetic and Evolutionary Computa- tion Conference

    Diggelen, F., Luo, J., Karag¨ uzel, T.A., Cambier, N., Ferrante, E., Eiben, A.E.: Environment induced emergence of collective behavior in evolving swarms with limited sensing. In: Proceedings of the Genetic and Evolutionary Computa- tion Conference. GECCO ’22, pp. 31–39. Assoc...

    2022

  45. [53]

    Evolutionary computation 9(4), 495–524 (2001)

    Urzelai, J., Floreano, D.: Evolution of adaptive synapses: Robots with fast adap- tive behavior in new environments. Evolutionary computation 9(4), 495–524 (2001)

    2001

  46. [54]

    arXiv preprint arXiv:2503.12406 (2025)

    Leung, B., Haomachai, W., Pedersen, J.W., Risi, S., Manoonpong, P.: Bio- inspired plastic neural networks for zero-shot out-of-distribution generalization in complex animal-inspired robots. arXiv preprint arXiv:2503.12406 (2025)

    2025 arXiv

  47. [55]

    Nature Neuroscience, 1–9 (2025)

    Miconi, T., Kay, K.: Neural mechanisms of relational learning and fast knowledge reassembly in plastic neural networks. Nature Neuroscience, 1–9 (2025)

    2025

  48. [56]

    arXiv preprint arXiv:2104.11523 (2021)

    Taffanel, A., Rousselot, B., Danielsson, J., McGuire, K., Richardsson, K., Elias- son, M., Antonsson, T., H¨ onig, W.: Lighthouse positioning system: dataset, accuracy, and precision for uav research. arXiv preprint arXiv:2104.11523 (2021)

    2021 arXiv

  49. [57]

    In: Proceedings of the 1999 Congress on Evolution- ary Computation-CEC99 (Cat

    Clerc, M.: The swarm and the queen: towards a deterministic and adaptive particle swarm optimization. In: Proceedings of the 1999 Congress on Evolution- ary Computation-CEC99 (Cat. No. 99TH8406), vol. 3, pp. 1951–19573 (1999). https://doi.org/10.1109/CEC.1999.785513 24 Appendi...

    1999

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.