Robots that can adapt like animals
Open accessThe DOI matches its Crossref record; 39 of the 40 references it cites were found in a registry or an index.
Abstract
Robots have transformed many industries, most notably manufacturing, and have the power to deliver tremendous benefits to society, such as in search and rescue, disaster response, health care and transportation. They are also invaluable tools for scientific exploration in environments inaccessible to humans, from distant planets to deep oceans. A major obstacle to their widespread adoption in more complex environments outside factories is their fragility. Whereas animals can quickly adapt to injuries, current robots cannot 'think outside the box' to find a compensatory behaviour when they are damaged: they are limited to their pre-specified self-sensing abilities, can diagnose only anticipated failure modes, and require a pre-programmed contingency plan for every type of potential damage, an impracticality for complex robots. A promising approach to reducing robot fragility involves having robots learn appropriate behaviours in response to damage, but current techniques are slow even with small, constrained search spaces. Here we introduce an intelligent trial-and-error algorithm that allows robots to adapt to damage in less than two minutes in large search spaces without requiring self-diagnosis or pre-specified contingency plans. Before the robot is deployed, it uses a novel technique to create a detailed map of the space of high-performing behaviours. This map represents the robot's prior knowledge about what behaviours it can perform and their value. When the robot is damaged, it uses this prior knowledge to guide a trial-and-error learning algorithm that conducts intelligent experiments to rapidly discover a behaviour that compensates for the damage. Experiments reveal successful adaptations for a legged robot injured in five different ways, including damaged, broken, and missing legs, and for a robotic arm with joints broken in 14 different ways. This new algorithm will enable more robust, effective, autonomous robots, and may shed light on the principles that animals use to adapt to injury.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“We render images using the MuJoCo Physics Engine's [45] built-in renderer. This renderer is not intended to be photo-realistic, and physically plausible choices of textures and lighting are not needed.”
From Domain randomization for transferring deep neural networks from simulation to the real world (2017) · cites this paper as [45]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“under Quality-Diversity optimization (Pugh et al., 2016), the search space is divided into behavioral niches to evolve diverse solutions, a paradigm that challenged the dominant approach of training a single agent, as it proved more robust in real-world settings (Cully et al., 2015).”
From Dynamics of niche construction in adaptable populations evolving in diverse environments (2023) · cites this paper as Cully et al., 2015
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Similarly, extensive reliance on simulation and offline training further distances learning from the full variability and constraint structure of real sensorimotor interaction (Jakobi et al., 1995; Cully et al., 2015).”
From Toward Enactive Artificial Intelligence (2026) · cites this paper as Cully et al., 2015
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Gaussian Processes for Machine Learning
Rasmussen, C. E., & Williams, C. K. I. (2005) · doi:10.7551/mitpress/3206.001.0001 · Cited by 11,705
Corke, P., Roberts, J., Cunningham, J., & Hainsworth, D. (2008) · Springer Handbook of Robotics · doi:10.1007/978-3-540-30301-5_50 · Cited by 3,543
A survey of robot learning from demonstration
Argall, B. D., Chernova, S., Veloso, M., & Browning, B. (2009) · Robotics and Autonomous Systems · doi:10.1016/j.robot.2008.10.024 · Cited by 3,227
Reinforcement learning in robotics: A survey
Kober, J., Bagnell, J. A., & Peters, J. (2013) · The International Journal of Robotics Research · doi:10.1177/0278364913495721 · Cited by 3,160 · cited 9×
“The key insight is that, whereas current learning algorithms either start with no knowledge of the search space 17 or with minimal knowledge from a few human demonstrations 17,18 , animals better understand the space of possible behaviors and their value from previous experience 19 , enabling…”
BackgroundMethodResultInfluentialDiagnosis and Fault-Tolerant Control
Blanke, M., Kinnaert, M., Lunze, J., & Staroswiecki, M. (2003) · Springer eBooks · doi:10.1007/978-3-662-05344-7 · Cited by 2,789
Nonlinear Programming: Sequential Unconstrained Minimization Techniques
Fiacco, A. V., & McCormick, G. P. (1968) · Defense Technical Information Center (DTIC) · Cited by 2,262
Stanley: The robot that won the DARPA Grand Challenge
Thrun, S., Montemerlo, M., Dahlkamp, H., Stavens, D., Aron, A., Diebel, J., Fong, P., Gale, J., Halpenny, M., Hoffmann, G., et al. (2006) · Journal of Field Robotics · doi:10.1002/rob.20147 · Cited by 2,237 · cited 2×
“3C, undam-aged: 24 [16; 41] seconds, 3 [2; 5] physical trials, repaired: 29 [16; 82] seconds, 3.5 [2; 10] trials).”
BackgroundInfluentialBrochu, E., Cora, V. M., & de Freitas, N. (2010) · arXiv (Cornell University) · doi:10.48550/arxiv.1012.2599 · Cited by 2,011
Bayesian integration in sensorimotor learning
Körding, K. P., & Wolpert, D. M. (2004) · Nature · doi:10.1038/nature02169 · Cited by 2,004 · cited 1×
“…optimization procedure followed by our robot appears similar to the technique employed by humans when they optimize an unknown function 23 , and there is strong evidence that animal brains learn probability distributions, combine them with prior knowledge, and act as Bayesian optimizers 25,26 .”
MethodControl of mental activities by internal models in the cerebellum
Ito, M. (2008) · Nature Reviews Neuroscience · doi:10.1038/nrn2332 · Cited by 1,356 · cited 1×
“A final parallel is that the simulator and Gaussian process components of Intelligent Trial and Error are two forms of predictive models, which are known to exist in animals 29,12 .”
Research works printed but not linked: 2
- Siciliano, B. & Khatib, O. Springer Handbook of Robotics (Springer, 2008). · doi:10.1007/978-3-540-30301-5
- Kluger, J. & Lovell, J. Apollo 13 (Mariner Books, 2006).
Cited by
Items 201–300 of 1,007
Adami, C. (2015) · Nature · doi:10.1038/521426a · Cited by 46
Learning soft task priorities for control of redundant robots
Modugno, V., Neumann, G., Rueckert, E., Oriolo, G., Peters, J., & Ivaldi, S. (2016) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2016.7487137 · Cited by 38
“Furthermore, recent results showed that it is possible to achieve very fast performances in trial-and-error learning with derivative-free methods [19].”
BackgroundValue systems for developmental cognitive robotics: A survey
Merrick, K. E. (2016) · Cognitive Systems Research · doi:10.1016/j.cogsys.2016.08.001 · Cited by 41
“For example, work on behavioural diversity generation through reuse and past experience (Benureau and Oudeyer, 2016) to create robots that can adapt like animals (Cully et al., 2015) is complementary to developmental cognitive robotics.”
BackgroundUsing Parameterized Black-Box Priors to Scale Up Model-Based Policy Search for Robotics
Chatzilygeroudis, K. I., & Mouret, J.-B. (2018) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2018.8461083 · Cited by 50
Self-citationUltracompact Photonic Structure Design for Strong Light Confinement and Coupling Into Nanowaveguide
Turduev, M., Bor, E., Latifoğlu, Ç., Giden, İ. H., Hanay, Y. S., & Kurt, H. (2018) · Journal of Lightwave Technology · doi:10.1109/jlt.2018.2821361 · Cited by 44
“Researchers leveraged machine learning successfully in developing a selflearning robot, which can withstand failures [22], improving”
BackgroundKim, S., Coninx, A., & Doncieux, S. (2020) · Robotics and Autonomous Systems · doi:10.1016/j.robot.2020.103710 · Cited by 43
Are quality diversity algorithms better at generating stepping stones than objective-based search?
Gaier, A., Asteroth, A., & Mouret, J.-B. (2019) · Genetic and Evolutionary Computation Conference Companion (GECCO Companion) · doi:10.1145/3319619.3321897 · Cited by 49
“To illustrate the capabilities of QD in this problem we maintain the same representation and variation operators, but guide evolution using the QD algorithm MAP-Elites [1, 5].”
MethodSelf-citationSchilling, M., Konen, K., Ohl, F. W., & Korthals, T. (2020) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros45743.2020.9341754 · Cited by 44
“One exemption is given by an evolutionary approach by Cully et al. who proposed a similar hierarchical structure and explicitly introduced an intermediate behavioral space as a lower dimensional representation for behavioral switching [20].”
BackgroundAn overview of self-engineering systems
Brooks, S., & Roy, R. (2021) · Journal of Engineering Design · doi:10.1080/09544828.2021.1914323 · Cited by 41
“Figure 10 shows pictures from Cully et al. (2015a) of the damage introduced and results of two self-adapting robots tested (Cully et al. 2015a).”
BackgroundMethodBiologicalisation in manufacturing – Current state and future trends
Wegener, K., Damm, O., Harst, S., Ihlenfeldt, S., Monostori, L., Teti, R., Wertheim, R., & Byrne, G. (2023) · CIRP Annals · doi:10.1016/j.cirp.2023.05.005 · Cited by 38
Discovering Diverse Multi-Agent Strategic Behavior via Reward RandomizationPreprint
Tang, Z., Yu, C., Chen, B., Xu, H., Wang, X., Fang, F., Du, S. S., Wang, Y., & Wu, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.04564 · Cited by 54
“RPG is also inspired by the map-elite algorithm (Cully et al., 2015) from evolutionary learning community, which optimizes multiple objectives simultaneously for sufficiently diverse polices.”
BackgroundDifferentiable Quality DiversityPreprint
Fontaine, M. C., & Nikolaidis, S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.03894 · Cited by 48
“Quality diversity (QD) is a rapidly emerging field [7] with applications including procedural content generation [27], damage recovery in robotics [13, 43], efficient aerodynamic shape design [22], and scenario generation in human-robot interaction [16, 17].”
BackgroundMethodInfluentialLi, R., & Zou, Z. (2023) · Advanced Engineering Informatics · doi:10.1016/j.aei.2023.102140 · Cited by 37
Toward Autonomous Self‐Healing in Soft Robotics: A Review and Perspective for Future Research
Tabrizian, S. K., Terryn, S., & Vanderborght, B. (2025) · Advanced Intelligent Systems · doi:10.1002/aisy.202400790 · Cited by 43
Barsce, J. C., Palombarini, J. A., & Martínez, E. (2017) · CLEI electronic journal · doi:10.1109/clei.2017.8226439 · Cited by 44
“[41, 42]), where such layer is used to learn a policy for a damaged robot, updating the model after each iteration.”
Backgroundpyribs: A Bare-Bones Python Library for Quality Diversity Optimization
Tjanaka, B., Fontaine, M. C., Lee, D. H., Zhang, Y., Balam, N. R., Dennler, N. S., Garlanka, S. S., Klapsis, N. D., & Nikolaidis, S. (2023) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3583131.3590374 · Cited by 57
“For instance, a roboticist may develop diverse walking gaits so that their robot can adapt to different morphological considerations [15], while a video game designer may generate multiple video game levels so that players can experience various levels of difficulty [22, 28], and a chemist may create multiple viable d…”
BackgroundHow do Different Encodings Influence the Performance of the MAP-Elites Algorithm?
Tarapore, D., Clune, J., Cully, A., & Mouret, J.-B. (2016) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/2908812.2908875 · Cited by 44
“…state-of-the-art encodings for locomotion controllers: (1) for comparison, the original, directly encoded, parameterized open loop controller from [9], (2) a CPPN [25] that specifies the parameters of a network of non-linear oscillators (Central Pattern Generators) [14], (3) a CPPN that specifies…”
BackgroundMethodInfluentialSelf-citationA terrain-adaptive robot prototype designed for bumpy-surface exploration
Zhang, F., Yu, Y., Wang, Q., Zeng, X., & Niu, H. (2019) · Mechanism and Machine Theory · doi:10.1016/j.mechmachtheory.2019.07.008 · Cited by 37
Task space adaptation via the learning of gait controllers of magnetic soft millirobots
Demir, S. Ö., Çulha, U., Karacakol, A. C., Pena‐Francesch, A., Trimpe, S., & Sitti, M. (2021) · The International Journal of Robotics Research · doi:10.1177/02783649211021869 · Cited by 39
“…of our experimental approach to alternative optimization and control methods supported with simulations such as intelligent trial and error (Cully et al., 2015), evolution algorithms (Kriegman et al., 2020), or policy gradients (Sehnke et al., 2010) is beyond the scope of our current study…”
MethodLearning directed locomotion in modular robots with evolvable morphologies
Lan, G., Carlo, M. D., van Diggelen, F., Tomczak, J. M., Roijers, D. M., & Eiben, A. E. (2021) · Applied Soft Computing · doi:10.1016/j.asoc.2021.107688 · Cited by 62
Homchanthanakul, J., & Manoonpong, P. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3119127 · Cited by 41
“Based on machine learning, different techniques have been proposed, including reinforcement learning [23], [24], intelligent trial and error [25], and evolutionary computation [26], [27].”
MethodOsa, T., Tangkaratt, V., & Sugiyama, M. (2022) · Neural Networks · doi:10.1016/j.neunet.2022.04.009 · Cited by 45
Asking better questions: How presentation formats influence information search.
Wu, C. M., Meder, B., Filimon, F., & Nelson, J. D. (2017) · Journal of Experimental Psychology Learning Memory and Cognition · doi:10.1037/xlm0000374 · Cited by 38
Advances of LiCoO2 in Cathode of Aqueous Lithium‐Ion Batteries
Ma, H., Wang, F., Shen, M., Tong, Y., Wang, H., & Hu, H. (2023) · Small Methods · doi:10.1002/smtd.202300820 · Cited by 38
Open-endedness in synthetic biology: A route to continual innovation for biological design
Stock, M., & Gorochowski, T. E. (2024) · Science Advances · doi:10.1126/sciadv.adi3621 · Cited by 38
“The MAP-elites algorithm has been found to be highly successful in evolutionary robotics ( 128 ).”
Dasgupta, S., Goldschmidt, D., Wörgötter, F., & Manoonpong, P. (2015) · Frontiers in Neurorobotics · doi:10.3389/fnbot.2015.00010 · Cited by 35
“…(as demonstrated with the gap crossing, climbing and uneven terrain navigation examples) but can also require the ability to adapt to sudden or abrupt changes in body properties, like growth or lesions (e.g., damage to robot joint motors or connections being disengaged) (Cully et al., 2015).”
BackgroundRegenerating Soft Robots Through Neural Cellular Automata
Horibe, K., Walker, K., & Risi, S. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-72812-0_3 · Cited by 53
CrowdHMT: Crowd Intelligence With the Deep Fusion of Human, Machine, and IoT
Guo, B., Liu, Y., Liu, S., Yu, Z., & Zhou, X. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3194726 · Cited by 38
“[67] proposed an intelligent trial-and-error algorithm inspired by animal adaptive mechanisms, which allows robots to find an adaptive solution in a short time”
BackgroundBayesian Optimization for Online Management in Dynamic Mobile Edge Computing
Yan, J., Lu, Q., & Giannakis, G. B. (2023) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2023.3307875 · Cited by 36
“Apart from the applications such as hyperparameter tuning in machine learning [29], drug discovery [30], and robotics [31], BO has been applied to several problems in the context of wireless networks, including radio resource allocation [32], coverage and capacity optimization in cellular networks…”
MethodBioinspired robots can foster nature conservation
Chellapurath, M., Khandelwal, P. C., & Schulz, A. (2023) · Frontiers in Robotics and AI · doi:10.3389/frobt.2023.1145798 · Cited by 36
“Adaptive structures, which can vary in shape and/or stiffness, can better withstand changing environmental conditions and tasks, reducing the stresses on the body (Cully et al., 2015; Khaheshi and Rajabi, 2022).”
Online Knowledge Level Tracking with Data-Driven Student Models and Collaborative Filtering
Cully, A., & Demiris, Y. (2019) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2019.2912367 · Cited by 34
“For instance, the mean term has been replaced by a ”behavioural map” in [56] to provide to a robot with prior knowledge about the different behaviours”
BackgroundSelf-citationFinding Game Levels with the Right Difficulty in a Few Trials through Intelligent Trial-and-Error
González-Duque, M., Palm, R. B., Ha, D., & Risi, S. (2020) · IEEE Conference on Games (CoG) · doi:10.1109/cog47356.2020.9231548 · Cited by 41
“In this article we tested whether the Intelligent Trial-andError algorithm (IT&E) [1] could be used for Dynamic Difficulty Adjustment (DDA); for this, we evolved several archives of difficult levels for different planning agents and lay them in a behavior map using MAP-Elites [30], and we used these maps as priors in…”
BackgroundMethodInfluentialPolykretis, I., Tang, G., & Michmizos, K. P. (2020) · International Conference on Neuromorphic Systems (ICONS) · doi:10.1145/3407197.3407205 · Cited by 36
“noisier and more complex scenarios [10].”
Fontaine, M. C., & Nikolaidis, S. (2021) · Robotics: Science and Systems · doi:10.15607/rss.2021.xvii.036 · Cited by 44
Versatile modular neural locomotion control with fast learning
Thor, M., & Manoonpong, P. (2022) · Nature Machine Intelligence · doi:10.1038/s42256-022-00444-0 · Cited by 34
“Unlike wheeled and flying robots, legged robots often have a redundant number of legs and can, with the right control, continue to move even with several leg impairments ( 3–5 ).”
Online Multimodal Ensemble Learning Using Self-Learned Sensorimotor Representations
Zambelli, M., & Demiris, Y. (2016) · IEEE Transactions on Cognitive and Developmental Systems · doi:10.1109/tcds.2016.2624705 · Cited by 38
“Because the internal models are learned by the robot continuously, they can handle changes in the robot morphology or in the robot sensory system, while eliminating the need for explicit analytical model formulation and dealing with model drifts [4].”
BackgroundHuman irrationality: both bad and good for reward inferencePreprint
Chan, L., Critch, A., & Dragan, A. D. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.06956 · Cited by 34
“In addition to the fact that real people are not perfectly rational, the reason we focus on inference is that imitation is not enough in some cases, e.g. if we need to behave in new environments [33, 9, 18], or transfer the policy to a robot with different dynamics [7, 14, 26].”
Adaptive Prior Selection for Repertoire-Based Online Adaptation in Robotics
Kaushik, R., Desreumaux, P., & Mouret, J.-B. (2020) · Frontiers in Robotics and AI · doi:10.3389/frobt.2019.00151 · Cited by 40
“These 3× 12 = 36 parameters define the elementary policy for the robot (see [12] for more details about this controller).”
BackgroundMethodInfluentialSelf-citationResilient help to switch and overlap hierarchical subsystems in a small human group
Fujii, K., Yokoyama, K., KOYAMA, T., Rikukawa, A., Yamada, H., & Yamamoto, Y. (2016) · Scientific Reports · doi:10.1038/srep23911 · Cited by 37
Policy manifold search: exploring the manifold hypothesis for diversity-based neuroevolution
Rakicevic, N., Cully, A., & Kormushev, P. (2021) · Proceedings of the Genetic and Evolutionary Computation Conference · doi:10.1145/3449639.3459320 · Cited by 32
“Each iteration potentially adds new policies to the collection, following the MAP-Elites framework [11], which are then used to refine the latent representation in the next iteration.”
BackgroundMethodInfluentialSelf-citationQin, H., Xiang, Y., Liu, F., Han, Y., & Wang, Y. (2025) · Swarm and Evolutionary Computation · doi:10.1016/j.swevo.2025.101849 · Cited by 33
Corucci, F., Cheney, N., Kriegman, S., Bongard, J., & Laschi, C. (2017) · Frontiers in Robotics and AI · doi:10.3389/frobt.2017.00034 · Cited by 34
A brittle star-like robot capable of immediately adapting to unexpected physical damage
Kano, T., Sato, E., Ono, T., Aonuma, H., Matsuzaka, Y., & Ishiguro, A. (2017) · Royal Society Open Science · doi:10.1098/rsos.171200 · Cited by 38
“Engineers have attempted to overcome this problem using several techniques such as learning [9–15] and trial-anderror methods [16].”
MethodChen, G., Jin, B., & Chen, Y. (2019) · Applied Mathematical Modelling · doi:10.1016/j.apm.2019.09.021 · Cited by 32
Grounded action transformation for sim-to-real reinforcement learning
Hanna, J. P., Desai, S., Karnan, H., Warnell, G., & Stone, P. (2021) · Machine Learning · doi:10.1007/s10994-021-05982-z · Cited by 46
“Cully et al. (2015) use a simulator to estimate fitness values for low-dimensional robot behaviors which gives the robot prior knowledge of how to adapt its behavior if it becomes damaged during real world operation.”
Animal play and evolution: Seven timely research issues about enigmatic phenomena
Burghardt, G. M., Pellis, S. M., Schank, J. C., Smaldino, P. E., Vanderschuren, L. J. M. J., & Palagi, E. (2024) · Neuroscience & Biobehavioral Reviews · doi:10.1016/j.neubiorev.2024.105617 · Cited by 38
Multimodal representation models for prediction and control from partial information
Zambelli, M., Cully, A., & Demiris, Y. (2019) · Robotics and Autonomous Systems · doi:10.1016/j.robot.2019.103312 · Cited by 39
Self-citationMultimodal soft valve enables physical responsiveness for preemptive resilience of soft robots
Pontin, M., & Damian, D. D. (2024) · Science Robotics · doi:10.1126/scirobotics.adk9978 · Cited by 34
The Many AI Challenges of Hearthstone
Hoover, A. K., Togelius, J., Lee, S., & Silva, F. D. M. (2019) · Künstliche Intell. · doi:10.1007/s13218-019-00615-z · Cited by 38
“h higher x-values and lower yvalues tend to perform best, gure 1b shows a variety of concentrations of highest-performing decks found with an algorithm that promotes deck diversity. Called MAP-Elites [10, 40], the algorithm builds a map of the best individuals and stores them if they also have unique behavior vectors.…”
BackgroundMulti-Robot Coordination and Layout Design for Automated Warehousing
Zhang, Y., Fontaine, M. C., Bhatt, V., Nikolaidis, S., & Li, J. (2023) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2023/611 · Cited by 31
“MAP-Elites [Cully et al., 2015; Mouret and Clune, 2015] is a popular QD algorithm.”
BackgroundEvolving parsimonious networks by mixing activation functions
Hagg, A., Mensing, M., & Asteroth, A. (2017) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3071178.3071275 · Cited by 31
“3071275 prone to continuous changes and novelty [3] or when using machine learning models to replace expensive sampling methods in optimization [6].”
MethodHart, E., & Goff, L. K. L. (2021) · Philosophical Transactions of the Royal Society B Biological Sciences · doi:10.1098/rstb.2021.0117 · Cited by 35
“However, there is an emerging body of work that takes morphological change into account when adapting behaviour: Cully et al. [50] considered the role of learning in the context of adapting controllers in response to morphological change that occurs during a lifetime, while Walker et al. [51] study…”
BackgroundRevolve: A Versatile Simulator for Online Robot Evolution
Hupkes, E., Jelisavcic, M., & Eiben, A. E. (2018) · Lecture notes in computer science · doi:10.1007/978-3-319-77538-8_46 · Cited by 36
“More complicated approaches may involve alternating fitness evaluations between simulation and reality [11,12,13,14], but this is infeasible when simulating entire artificial ecosystems.”
BackgroundFrom privacy to anti-discrimination in times of machine learning
Hagendorff, T. (2019) · Ethics and Information Technology · doi:10.1007/s10676-019-09510-5 · Cited by 32
“Machine learning is applied for purposes of speech or image recognition, computer vision, recommendation systems, language processing, spam recognition, ranking systems, robot or autonomous vehicle control, personality analysis, automatic tasks in fields like law, finance, medicine and so on (Sajjadi et al. 2017; Silv…”
BackgroundPlanning in Learned Latent Action Spaces for Generalizable Legged Locomotion
Li, T., Calandra, R., Pathak, D., Tian, Y., Meier, F., & Rai, A. (2021) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2021.3062342 · Cited by 38
“[14], [15].”
Deep surrogate assisted MAP-elites for automated hearthstone deckbuilding
Zhang, Y., Fontaine, M. C., Hoover, A. K., & Nikolaidis, S. (2022) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3512290.3528718 · Cited by 48
“As a result, QD algorithms have been applied to procedural content generation [40], robotics [13, 33], aerodynamic shape design [11], and scenario generation in human-robot interaction [41, 42].”
MethodFlageat, M., & Cully, A. (2023) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2023.3273560 · Cited by 37
“been applied to a wide range of domains ranging from robotics [7], [8] to content generation [5], [9] or design [4].”
BackgroundSelf-citationEvolutionary multiobjective design of a flexible caudal fin for robotic fish
Clark, A. J., Tan, X., & McKinley, P. K. (2015) · Bioinspiration & Biomimetics · doi:10.1088/1748-3190/10/6/065006 · Cited by 31
Behavioral Diversity Generation in Autonomous Exploration through Reuse of Past Experience
Benureau, F., & Oudeyer, P. (2016) · Frontiers in Robotics and AI · doi:10.3389/frobt.2016.00008 · Cited by 32
“…2012), and that related algorithms implemented as part of the SAGG-Random architecture such as in Baranes and Oudeyer (2013) and Moulin-Frier et al. (2014), as well as other variations such as the MAP-Elite algorithm (Cully et al., 2015), have yielded good results in diverse robotics contexts.”
MethodInfluentialStanton, C., & Clune, J. (2016) · PLOS ONE · doi:10.1371/journal.pone.0162235 · Cited by 27
“Evolutionary Algorithms (EAs) have proven useful in a variety of fields, including space exploration [1], automated damage recovery [2], and evolving gaits for robots [3, 4].”
BackgroundMethodSelf-citationPlasticity of Sensorimotor Networks: Multiple Overlapping Mechanisms
Buch, E. R., Liew, S., & Cohen, L. G. (2016) · The Neuroscientist · doi:10.1177/1073858416638641 · Cited by 32
NoRML: No-Reward Meta Learning
Yang, Y., Caluwaerts, K., Işçen, A., Tan, J., & Finn, C. (2019) · Adaptive Agents and Multi-Agents Systems · doi:10.65109/elvr9404 · Cited by 32
Li, L., Shi, Z., Liang, H., Liu, J., & Qiao, Z. (2022) · Nanomaterials · doi:10.3390/nano12010159 · Cited by 30
“In recent years, ML has been widely applied to various fields, including image recognition [21], natural language processing [22], data classification and mining [23,24], and material performance prediction [25].”
BackgroundEmpirical analysis of PGA-MAP-Elites for Neuroevolution in Uncertain Domains
Flageat, M., Chalumeau, F., & Cully, A. (2023) · ACM Transactions on Evolutionary Learning and Optimization · doi:10.1145/3577203 · Cited by 38
“N E ] 2 4 O ct 2 02 2 111:2 Manon Flageat, Felix Chalumeau, and Antoine Cully aerodynamic design [Gaier et al. 2017] to robotics [Cully et al. 2015].”
BackgroundMethodInfluentialSelf-citationSatisficing in Multi-Armed Bandit Problems
Reverdy, P. B., Srivastava, V., & Leonard, N. E. (2016) · IEEE Transactions on Automatic Control · doi:10.1109/tac.2016.2644380 · Cited by 48
“In [10], the authors used a multi-armed bandit algorithm to construct robots that actively adapt their control policies to mitigate damage,”
MethodAerodynamic Design Exploration through Surrogate-Assisted Illumination
Gaier, A., Asteroth, A., & Mouret, J.-B. (2017) · AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference (MAO) · doi:10.2514/6.2017-3330 · Cited by 40
Self-citationObject carrying of hexapod robots with integrated mechanism of leg and arm
Deng, H., Xin, G., Zhong, G., & Mistry, M. (2017) · Robotics and Computer-Integrated Manufacturing · doi:10.1016/j.rcim.2017.11.014 · Cited by 32
“That has been demonstrated by researchers who do research on fault tolerant gaits [25-27].”
Urquhart, N. B., & Hart, E. (2018) · Lecture notes in computer science · doi:10.1007/978-3-319-99253-2_39 · Cited by 29
Preparation of ordered states in ultra-cold gases using Bayesian optimization
Mukherjee, R., Sauvage, F., Xie, H., Löw, R., & Mintert, F. (2020) · New Journal of Physics · doi:10.1088/1367-2630/ab8677 · Cited by 31
Acceleration for Deep Reinforcement Learning using Parallel and Distributed Computing: A Survey
Liu, Z., Xu, X., Qiao, P., & Li, D. (2024) · ACM Computing Surveys · doi:10.1145/3703453 · Cited by 39
“Some examples incorporate quality diversity [158, 159], novelty search [160, 161] and adaptive noise [162, 163] to improve exploration.”
Schilling, M., Melnik, A., Ohl, F. W., Ritter, H. J., & Hammer, B. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.09.017 · Cited by 33
Online Kinematic Calibration for Legged Robots
Yang, S., Choset, H., & Manchester, Z. R. (2022) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2022.3186501 · Cited by 28
OFFER: Off-Environment Reinforcement Learning
Ciosek, K., & Whiteson, S. (2017) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v31i1.10810 · Cited by 49
“An off-theshelf kinematics simulator (Cully et al. 2015) models an arm with three joints, each of which can be set by the agent, whose goal is to get the end effector to a specific location.”
BackgroundThe trade-off between morphology and control in the co-optimized design of robots
Rosendo, A., von Atzigen, M., & Iida, F. (2017) · PLoS ONE · doi:10.1371/journal.pone.0186107 · Cited by 34
Towards the targeted environment-specific evolution of robot components
Collins, J., Geles, W., Howard, D., & Maire, F. (2018) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3205455.3205541 · Cited by 28
“ots display decreased performance when physically created, which is known as crossing the reality gap [18]. Mixing simulation and reality is a promising area of research, through e.g. internal models [6, 9], however it is unknown if model fidelity can match complex real-world environmental conditions. We cannot feasib…”
BackgroundContinuous learning of emergent behavior in robotic matter
Oliveri, G., van Laake, L. C., Carissimo, C., Miette, C., & Overvelde, J. T. B. (2021) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2017015118 · Cited by 29
“By sensing their environment, robots can build models of themselves (4) or optimize human-made models with machine learning strategies to improve their behavior (1, 2, 4–8).”
Chen, Y., Yang, C., Gu, Y., & Hu, B. (2022) · IEEE Transactions on Human-Machine Systems · doi:10.1109/thms.2021.3134553 · Cited by 33
AutoPipe: A Fast Pipeline Parallelism Approach with Balanced Partitioning and Micro-batch Slicing
Liu, W., Lai, Z., Li, S., Duan, Y., Ge, K., & Li, D. (2022) · IEEE International Conference on Cluster Computing (CLUSTER) · doi:10.1109/cluster51413.2022.00042 · Cited by 37
“Deep neural networks (DNNs) have been successfully advanced in many domains [1]–[3].”
BackgroundLimping following limb loss increases locomotor stability
Wilshin, S. D., Shamble, P. S., Hovey, K. J., Harris, R., Spence, A. J., & Hsieh, S. T. (2018) · Journal of Experimental Biology · doi:10.1242/jeb.174268 · Cited by 27
“…resilience displayed by wolf spiders should therefore be Jo u rn al o f Ex p er im en ta lB io lo g y of interest to engineers and roboticists, providing a simple solution for combatting destabilizing perturbations that are common during natural, untethered locomotion (Cully et al., 2015).”
BackgroundLow-Cost Wireless Modular Soft Tensegrity Robots
Kimber, J., Ji, Z., Petridou, A., Sipple, T., Barhydt, K., Boggs, J., Dosiek, L., & Rieffel, J. A. (2019) · IEEE International Conference on Soft Robotics (RoboSoft) · doi:10.1109/robosoft.2019.8722723 · Cited by 27
“which means that the development of successful locomotive gaits will need to rely upon intelligent trial and error methods (such as Bayesian Optimization [27]) which can find results while minimizing the number of physical trials.”
BackgroundMethodInfluentialApplication cases of biological transformation in manufacturing technology
Bergs, T., Schwaneberg, U., Barth, S., Hermann, L., Grunwald, T., Mayer, S., Biermann, F., & Sözer, N. (2020) · CIRP journal of manufacturing science and technology · doi:10.1016/j.cirpj.2020.09.010 · Cited by 28
Diverse Auto-Curriculum is Critical for Successful Real-World Multiagent Learning Systems
Yang, Y., Luo, J., Wen, Y., Slumbers, O., Graves, D., Ammar, H. B., Wang, Y.-X., & Taylor, M. E. (2021) · Adaptive Agents and Multi-Agents Systems · doi:10.65109/lixm5646 · Cited by 38
MAP-Elites with Descriptor-Conditioned Gradients and Archive Distillation into a Single Policy
Faldor, M., Chalumeau, F., Flageat, M., & Cully, A. (2023) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3583131.3590503 · Cited by 40
“Producing a large collection of diverse and effective solutions enables to get multiple alternatives to solve a single problem which is useful in robotics to improve robustness, recover from damage [6] or reduce the reality gap [4].”
BackgroundSelf-citationBootstrapping $Q$ -Learning for Robotics From Neuro-Evolution Results
Zimmer, M., & Doncieux, S. (2017) · IEEE Transactions on Cognitive and Developmental Systems · doi:10.1109/tcds.2016.2628817 · Cited by 29
“If most work are focused on locomotion in a straight line, recent work have defined a control architecture including evolutionary approaches and able to learn to walk in any direction [57], and even to adapt to motor failures [58].”
Deep Kernels for Optimizing Locomotion ControllersPreprint
Antonova, R., Rai, A., & Atkeson, C. G. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.09062 · Cited by 47
“[3] tabulate best performing points versus their average score on a behavioural metric – average contact time of their hexapod system in simulation.”
BackgroundMethodInfluentialWhy Space Migration Must Be Posthuman
Ferrando, F. (2016) · Space and society · doi:10.1007/978-3-319-39827-3_10 · Cited by 30
Miguel-Blanco, A., & Manoonpong, P. (2020) · Frontiers in Neural Circuits · doi:10.3389/fncir.2020.00046 · Cited by 32
“, 2016) or machine learning techniques (Bongard et al., 2006; Cully et al., 2015; Hwangbo et al., 2019), or their combination (like CPGbased control with reinforcement learning, Ishige et al.”
BackgroundAdaptive Robotic Information Gathering via non-stationary Gaussian processes
Chen, W., Khardon, R., & Liu, L. (2023) · The International Journal of Robotics Research · doi:10.1177/02783649231184498 · Cited by 33
“Due to the complex interaction between the robot and the environment, e.g. , a quadruped running at high speed over rough terrain, mechanical wear and tear, and actuator faults, it may be infeasible to build an accurate dynamics model a priori (Cully et al. 2015).”
BackgroundAn ontological framework for organising and describing behaviours: The Human Behaviour Ontology
Schenk, P. M., West, R., Castro, Ó., Hayes, E., Hastings, J., Johnston, M., Marques, M. M., Corker, E., Wright, A. J., Stuart, G., et al. (2024) · Wellcome Open Research · doi:10.12688/wellcomeopenres.21252.1 · Cited by 27
Learning an Optimization Algorithm Through Human Design Iterations
Sexton, R., & Ren, M. Y. (2017) · Journal of Mechanical Design · doi:10.1115/1.4037344 · Cited by 25
“In fact, many higher-dimensional problems that human beings naturally solve, such as locomotion planning, have also been successfully solved through the use of GP [19–22].”
BackgroundA Performance Oriented Novel Design of Hexapod Robots
Zhong, G., Chen, L., & Deng, H. (2017) · IEEE/ASME Transactions on Mechatronics · doi:10.1109/tmech.2017.2681722 · Cited by 25
“Thus, the fault tolerance [33]–[35] is very important to develop an excellent robot.”
Background1D Printing of Recyclable Robots
Cellucci, D., MacCurdy, R. B., Lipson, H., & Risi, S. (2017) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2017.2716418 · Cited by 31
Tanneberg, D., Peters, J., & Rueckert, E. (2018) · Neural Networks · doi:10.1016/j.neunet.2018.10.005 · Cited by 29
“…spaces and settings with differ-ent modalities was shown – and learning the state neuron population [62], in order to apply the framework to recover from failure tasks with broken joints [66, 67], investigating an intrinsic motivation signal mimicking the avoidance of arthritic pain [68, 69].”
BackgroundRobust optimization through neuroevolution
Pagliuca, P., & Nolfi, S. (2019) · PLoS ONE · doi:10.1371/journal.pone.0213193 · Cited by 30
Bayesian Optimization Meets Riemannian Manifolds in Robot LearningPreprint
Jaquier, N., Rozo, L. D., Calinon, S., & Bürger, M. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1910.04998 · Cited by 25
“A recent approach to improve BO performance is via domain knowledge, commonly introduced into the GP mean function [2] or through the design of task-specific kernels [10], as detailed in § 2.”
BackgroundInfluentialBayesian Optimization for Quality Diversity Search With Coupled Descriptor Functions
Kent, P., Gaier, A., Mouret, J.-B., & Branke, J. (2024) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2024.3376733 · Cited by 30
“This introduces the possibility of region misprediction, i.e., predicting that a point maps to a certain region which, when evaluated on the true descriptor functions, turns out to be incorrect.”
InfluentialSelf-citationNeugebauer, R., Ihlenfeldt, S., Schließmann, U., Hellmich, A., & NOACK, M. (2019) · Journal of Machine Engineering · doi:10.5604/01.3001.0013.0440 · Cited by 24
“a) A dog relearns to run with three legs, b) Undamaged hexapod robot, c) A possible damage of the hexapod robot, d) Hexapod robot learns to walk with five legs [13]”
Learning of Sub-optimal Gait Controllers for Magnetic Walking Soft Millirobots
Çulha, U., Demir, S. Ö., Trimpe, S., & Sitti, M. (2020) · Robotics: Science and Systems · doi:10.15607/rss.2020.xvi.070 · Cited by 27
“The systematic comparison of our experimental approach to alternative methods supported with simulations such as intelligent trial and error [8], evolution algorithms [20], or policy gradients [33] is also an interesting future work.”
ResultSensing flow gradients is necessary for learning autonomous underwater navigation
Jiao, Y., Hang, H., Merel, J., & Kanso, E. (2025) · Nature Communications · doi:10.1038/s41467-025-58125-6 · Cited by 23
The Convergence of AI and animal-inspired robots for ecological conservation
Afzal, N., Rehman, M. U., Seneviratne, L. D., & Hussain, I. (2024) · Ecological Informatics · doi:10.1016/j.ecoinf.2024.102950 · Cited by 26
Metadata
- Authors
- Cully, A., Clune, J., Tarapore, D., & Mouret, J.-B.
- Affiliations
- Centre National de la Recherche Scientifique (FR)
- University of Wyoming (US)
- Corresponding author
- Mouret, Jean-Baptiste
- Venue
- Nature
- Published
- 2015-05-27
- Volume
- 521
- Issue
- 7553
- Pages
- 503-507
- ISSN
- 0028-0836 (ISSN-L)
- 1476-4687
- OpenAlex
- W1738827650
- Cited by
- 979
- References
- 40
- Citations by year
- 2026: 51 · 2025: 76 · 2024: 88 · 2023: 90 · 2022: 100 · 2021: 115 · 2020: 129 · 2019: 112 · 2018: 78 · 2017: 78 · 2016: 48 · 2015: 13
- FWCI
- 70.47
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- www.springer.com/tdm (tdm)
- www.springer.com/tdm (vor)
- Locations
- Nature (publishedVersion)
- PubMed (publishedVersion)
- ePrints Soton (University of Southampton) (acceptedVersion)
- arXiv (Cornell University) (submittedVersion)
- HAL (submittedVersion, other-oa)
- RePEc: Research Papers in Economics (submittedVersion)
- arXiv (Cornell University)
- Funders
- Agence Nationale de la Recherche (ANR-12-JS03-0009)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Physical Sciences › Engineering › Mechanical Engineering › Modular Robots and Swarm Intelligence
- Physical Sciences › Computer Science › Artificial Intelligence › Evolutionary Algorithms and Applications
- Keywords
- compensatory behavior, trial-and-error learning, autonomous mobile robots, legged robots, robot manipulators, prior knowledge, robot robustness, search and rescue robots, disaster response
- MeSH terms
- Adaptation, Physiological, Algorithms, Animals, Artificial Intelligence, Behavior, Animal, Dogs, Extremities, Motor Skills, Robotics, Time Factors, Biomimetics
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.