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 101–200 of 1,007
Zhang, Y., Jiang, S., Lin, J., & Lee, C. (2020) · Journal of Materials Chemistry C · doi:10.1039/c9tc05758g · Cited by 76
Curiosity Driven Exploration of Learned Disentangled Goal SpacesPreprint
Laversanne-Finot, A., Péré, A., & Oudeyer, P. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1807.01521 · Cited by 86
“This is key for acquiring world models and repertoires of parameterized skills [3, 4, 5], to efficiently bootstrap exploration for deep reinforcement learning problems with rare or deceptive rewards [6, 7], or to quickly repair strategies in case of damages [8].”
BackgroundEvolution Through Large Models
Lehman, J., Gordon, J., Jain, S., Ndousse, K., Yeh, C., & Stanley, K. O. (2024) · Genetic and Evolutionary Computation · doi:10.1007/978-981-99-3814-8_11 · Cited by 74
Scaling MAP-Elites to deep neuroevolution
Colas, C., Madhavan, V., Huizinga, J., & Clune, J. (2020) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3377930.3390217 · Cited by 103
“[13] and use a Matérn kernel function with parameters ν = 5/2 and ρ = 0.”
BackgroundMethodInfluentialSelf-citationEvolution through Large Models
Lehman, J., Gordon, J., Jain, S., Ndousse, K., Yeh, C., & Stanley, K. O. (2022) · arXiv · doi:10.48550/arxiv.2206.08896 · Cited by 74
Locomotion Control and Gait Planning of a Novel Hexapod Robot Using Biomimetic Neurons
Zhong, G., Chen, L., Jiao, Z., Li, J., & Deng, H. (2017) · IEEE Transactions on Control Systems Technology · doi:10.1109/tcst.2017.2692727 · Cited by 73
“Besides, hexapod robots can use the remained legs for continuous work when one or two legs break down [2].”
BackgroundRapidly Adaptable Legged Robots via Evolutionary Meta-Learning
Song, X., Yang, Y., Choromański, K., Caluwaerts, K., Gao, W., Finn, C., & Tan, J. (2020) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros45743.2020.9341571 · Cited by 102
“Prior work has investigated online system identification [49], action transformation [15], precomputed behavior-performance map [5], strategy optimization [50, 51] and meta-learning [29] for quick adaptation of locomotion controllers.”
BackgroundMapping hearthstone deck spaces through MAP-elites with sliding boundaries
Fontaine, M. C., Lee, S., Soros, L. B., Silva, F. D. M., Togelius, J., & Hoover, A. K. (2019) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3321707.3321794 · Cited by 72
“tial impact, QD algorithms are predominately explored in only a fraction of the possible domains that may benefit from them, including problems in traditional evolutionary robotics such as locomotion [8] and maze navigation [38]. However, games offer a potentially fruitful avenue for QD research because of the multitu…”
BackgroundHuang, L., Zhu, Z., & Zou, Z. (2022) · Automation in Construction · doi:10.1016/j.autcon.2022.104691 · Cited by 78
Bayesian Optimization Using Domain Knowledge on the ATRIAS Biped
Rai, A., Antonova, R., Song, S., Martin, W., Geyer, H., & Atkeson, C. G. (2018) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2018.8461237 · Cited by 82
Advanced Design of Soft Robots with Artificial Intelligence
Cao, Y., Xu, B., Li, B., & Fu, H. C. (2024) · Nano-Micro Letters · doi:10.1007/s40820-024-01423-3 · Cited by 78
All‐Flexible Artificial Reflex Arc Based on Threshold‐Switching Memristor
Wang, D., Zhao, S., Li, L., Wang, L., Cui, S., Wang, S., Lou, Z., & Shen, G. (2022) · Advanced Functional Materials · doi:10.1002/adfm.202200241 · Cited by 71
Neuroadaptive Bayesian Optimization and Hypothesis Testing
Lorenz, R., Hampshire, A., & Leech, R. (2017) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2017.01.006 · Cited by 71
A Systematic Literature Review of the Successors of “NeuroEvolution of Augmenting Topologies”
Papavasileiou, E., Cornelis, J. P., & Jansen, B. (2020) · Evolutionary Computation · doi:10.1162/evco_a_00282 · Cited by 79
“In this article, MAP-Elites acts as the EA of the “Intelligent Trial & Error algorithm” (Cully et al., 2015) that allows robots to learn creative behaviors in order to adapt to unexpected situations such as damage.”
MethodRecent Advancements in Biomimetic 3D Printing Materials With Enhanced Mechanical Properties
Yan, X., Bethers, B., Chen, H., Xiao, S., Lin, S., Tran, B., Jiang, L., & Yang, Y. (2021) · Frontiers in Materials · doi:10.3389/fmats.2021.518886 · Cited by 67
Evolution of Repertoire-Based Control for Robots With Complex Locomotor Systems
Duarte, M., Gomes, J., Oliveira, S. M., & Christensen, A. L. (2017) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2017.2722101 · Cited by 67
“Unlike NSLC, the recent MAP-Elites algorithm [14], [17] divides the behavior space into discrete bins, and aims at finding high-quality solutions in each bin of the behavior space (the behavior repertoire).”
BackgroundMethodInfluentialRobot control parameters auto-tuning in trajectory tracking applications
Roveda, L., Forgione, M., & Piga, D. (2020) · Control Engineering Practice · doi:10.1016/j.conengprac.2020.104488 · Cited by 74
A Minimal Model Describing Hexapedal Interlimb Coordination: The Tegotae-Based Approach
Owaki, D., Goda, M., Miyazawa, S., & Ishiguro, A. (2017) · Frontiers in Neurorobotics · doi:10.3389/fnbot.2017.00029 · Cited by 77
“However, because our model did not include any directional or posture controls and learning algorithms as in Ren et al. (2015) and Cully et al. (2015) (here, we mainly focus on real-time adaptability), its direction of motion would vary according to the physical properties: a frontleft-leg…”
Method20 years of reality gap: a few thoughts about simulators in evolutionary robotics
Mouret, J.-B., & Chatzilygeroudis, K. I. (2017) · Genetic and Evolutionary Computation Conference Companion (GECCO Companion) · doi:10.1145/3067695.3082052 · Cited by 92
“…simulations were performed with an intact robot. reality gap was, therefore very important since we added the issues of the simulator to the damage; nonetheless, we were able to find high-performing gaits for all the damage conditions that we tested in only a dozen of tests on the real robot [4].”
BackgroundMethodSelf-citationGeneric Neural Locomotion Control Framework for Legged Robots
Thor, M., Kulvičius, T., & Manoonpong, P. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.3016523 · Cited by 71
“The main difference between the frameworks is that in [1] many locomotion control policies are learned at random and put into a behavior-performance map for handling many different types of damage to a legged robot, whereas in our”
BackgroundMethodInfluentialBing, Z., Lemke, C., Cheng, L., Huang, K., & Knoll, A. (2020) · Neural Networks · doi:10.1016/j.neunet.2020.05.029 · Cited by 69
“Their solution has proven that can significantly improve the performances of many robotic control tasks [36]. e In [37], two prototype experiments showed that RL methods could help robots recover from damage and adapt quickly like animals do.”
BackgroundPOET: open-ended coevolution of environments and their optimized solutions
Wang, R., Lehman, J., Clune, J., & Stanley, K. O. (2019) · Proceedings of the Genetic and Evolutionary Computation Conference · doi:10.1145/3321707.3321799 · Cited by 61
“Novelty search with local competition (NSLC) [37] and MAP-Elites [9, 42] rectify this limitation by adding a quality pressure (hence ushering in QD in earnest), but they still inevitably hit the limit of what is possible to discover in the particular environment where they are run–there are only so many novel ways a h…”
BackgroundMethodInfluentialSelf-citationSun, P., Guo, Z., Wang, G., Lan, J., & Hu, Y. (2020) · Computer Networks · doi:10.1016/j.comnet.2020.107230 · Cited by 69
Effective Diversity in Population-Based Reinforcement Learning
Parker-Holder, J., Pacchiano, A., Choromanski, K., & Roberts, S. (2020) · arXiv · doi:10.48550/arxiv.2002.00632 · Cited by 60
“This phenomenon motivates the MAP-Elites algorithm (Mouret and Clune, 2015; Cully et al., 2015), whereby only one solution may lie in each quadrant of a pre-defined space.”
BackgroundMethodTuring learning: a metric-free approach to inferring behavior and its application to swarms
Li, W., Gauci, M., & Groß, R. (2016) · Swarm Intelligence · doi:10.1007/s11721-016-0126-1 · Cited by 65
“…the implementation of evolution directly in physical environments, on either a single robot [Floreano and Mondada, 1996, Zykov et al., 2004, Bongard et al., 2006, Koos et al., 2013, Cully et al., 2015] or multiple robots [Watson et al., 2002, O’Dowd et al., 2014, Heinerman et al., 2015].”
BackgroundAutomated Shapeshifting for Function Recovery in Damaged Robots
Kriegman, S., Walker, S., Shah, D., Levin, M. G., Kramer-Bottiglio, R. K., & Bongard, J. (2019) · Robotics: Science and Systems · doi:10.15607/rss.2019.xv.028 · Cited by 74
“[13] proposed that robots should instead exploit the fact that resources prior to deployment are relatively cheap in terms of the factors listed above.”
BackgroundMethodInfluentialDu, L., Xu, Z., Huang, C.-L., Zhao, F.-Y., Fan, C., Dai, J., Yang, K., & Wang, Y. (2019) · Applied Materials Today · doi:10.1016/j.apmt.2019.100463 · Cited by 60
Adversarial Examples - A Complete Characterisation of the PhenomenonPreprint
Serban, A. C., Poll, E., & Visser, J. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1810.01185 · Cited by 70
Unscented Bayesian optimization for safe robot grasping
Nogueira, J., Martínez-Cantín, R., Bernardino, A. J. M., & Jamone, L. (2016) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros.2016.7759310 · Cited by 58
“In robotics in general and, for the problem of robot grasping in particular, Bayesian optimization has been one of the most successful and efficient trial-and-error techniques [2], [3], [4], [5], even in the presence of mechanical failures [6].”
BackgroundSparse Robot Swarms: Moving Swarms to Real-World Applications
Tarapore, D., Groß, R., & Zauner, K. (2020) · Frontiers in Robotics and AI · doi:10.3389/frobt.2020.00083 · Cited by 63
“Trial-and-error based algorithms for rapid behavior adaptation (e.g., see Cully et al., 2015) appear to be a promising approach to begin addressing these challenges.”
BackgroundMethodSelf-citationUsing Deep Reinforcement Learning to Learn High-Level Policies on the ATRIAS Biped
Li, T., Geyer, H., Atkeson, C. G., & Rai, A. (2019) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2019.8793864 · Cited by 65
“Recent work learns parameters of expert controllers on hardware sample-efficiently, for example [9] and [10] use Bayesian Optimization.”
MethodUnsupervised Behavior Discovery With Quality-Diversity Optimization
Grillotti, L., & Cully, A. (2022) · IEEE Transactions on Evolutionary Computation · doi:10.1109/tevc.2022.3159855 · Cited by 75
“These values for N C have the same order of magnitude as most resulting container sizes in the QD literature [6], [13]–[15], [19].”
BackgroundInfluentialSelf-citationGiordano, G., Carlotti, M., & Mazzolai, B. (2021) · Advanced Materials Technologies · doi:10.1002/admt.202100437 · Cited by 58
Data-Efficient Design Exploration through Surrogate-Assisted Illumination
Gaier, A., Asteroth, A., & Mouret, J.-B. (2018) · Evolutionary Computation · doi:10.1162/evco_a_00231 · Cited by 86
“Quality diversity (QD) algorithms (Pugh et al., 2016; Mouret and Clune, 2015; Cully et al., 2015) use evolutionary methods to produce a set of diverse, high-quality solutions within a single run.”
BackgroundMethodInfluentialSelf-citationDrug design targeting active posttranslational modification protein isoforms
Meng, F., Liang, Z. T., Zhao, K., & Luo, C. (2020) · Medicinal Research Reviews · doi:10.1002/med.21774 · Cited by 61
Exploring optimal control of epidemic spread using reinforcement learning
Ohi, A. Q., Mridha, M. F., Monowar, M. M., & Hamid, M. A. (2020) · Scientific Reports · doi:10.1038/s41598-020-79147-8 · Cited by 90
Biodiversity in evolved voxel-based soft robots
Medvet, E., Bartoli, A., Pigozzi, F., & Rochelli, M. (2021) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3449639.3459315 · Cited by 71
“Several challenges have to be tackled in order to achieve that goal, ranging from scalability in the design and building phases [16, 25], possibly with auto-fabrication, to effective mechanisms for robot adaptation to changes in the environment [7].”
BackgroundLevin, M. G. (2024) · Entropy · doi:10.3390/e26060481 · Cited by 74
“…of a robot that did not know its shape and had to discover it, and then re-use that information when its body shape was abruptly changed due to damage [235], and other work on the transfer of skills in dynamical recurrent neural networks [236] or deep learning in embodied robotics [237].”
BackgroundLearning Behavior Characterizations for Novelty Search
Meyerson, E., Lehman, J., & Miikkulainen, R. P. (2016) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/2908812.2908929 · Cited by 54
“Because it is di � cult to scale such objective-based search, there has been recent interest in diversity-driven algorithms [3, 13, 14, 19, 22, 26].”
BackgroundScaling Geo-Distributed Network Function Chains: A Prediction and Learning Framework
Luo, Z., Wu, C., Li, Z., & Zhou, W. (2019) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2019.2927068 · Cited by 56
“Atari [13]) and robot control [14].”
Discovering representations for black-box optimization
Gaier, A., Asteroth, A., & Mouret, J.-B. (2020) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3377930.3390221 · Cited by 68
“In robot locomotion, this could correspond to periodic functions, since we already know that a 36-dimensional controller based on periodic functions can produce the numerous joint commands required every second to effectively drive a 12-joint walking robot in many different ways [11].”
BackgroundMethodSelf-citationWang, R., Lehman, J., Rawal, A., Zhi, J., Li, Y., Clune, J., & Stanley, K. O. (2020) · arXiv · doi:10.48550/arxiv.2003.08536 · Cited by 53
“task if it looks promising for that other task) dramatically improves performance, including solving previously unsolvable problems like Montezuma’s Revenge or rapid damage recovery in robots (Cully et al., 2015; Ecoffet et al., 2019; Lehman & Stanley, 2011b; Mouret & Clune, 2015; Nguyen et al., 2016).”
BackgroundSelf-citationMuscle-inspired soft robots based on bilateral dielectric elastomer actuators
Yang, Y., Li, D., Sun, Y., Wu, M., Su, J., Li, Y., Yu, X., Li, L., & Yu, J. (2023) · Microsystems & Nanoengineering · doi:10.1038/s41378-023-00592-2 · Cited by 54
Evolutionary Developmental Robotics: Improving Morphology and Control of Physical Robots
Vujović, V., Rosendo, A., Brodbeck, L., & Iida, F. (2017) · Artificial Life · doi:10.1162/artl_a_00228 · Cited by 63
“Given the available computational power, most systems try to circumvent this problem with simulated environments, some of which are eventually transferred to the real-world [6, 4, 1].”
BackgroundLLMatic: Neural Architecture Search Via Large Language Models And Quality Diversity Optimization
Nasir, M. U., Earle, S., Togelius, J., James, S., & Cleghorn, C. (2024) · Proceedings of the Genetic and Evolutionary Computation Conference · doi:10.1145/3638529.3654017 · Cited by 52
“QD methods are valuable in domains such as robot control, where it is useful to learn diverse high-quality trajectories, in case one solution should become unavailable during deployment because of a physical obstruction or mechanical malfunction [9].”
MethodFast Online Adaptation in Robotics through Meta-Learning Embeddings of Simulated Priors
Kaushik, R., Anne, T., & Mouret, J.-B. (2020) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros45743.2020.9341462 · Cited by 77
“By contrast, we expect robots to adapt in seconds or, at worst, in minutes when they need to adapt to a new situation [4], [13].”
BackgroundSelf-citationAutonomous hydrogel locomotion regulated by light and electric fields
Yang, Y., Li, C., Palmer, L. C., & Stupp, S. I. (2023) · Science Advances · doi:10.1126/sciadv.adi4566 · Cited by 51
Data-efficient Learning of Morphology and Controller for a Microrobot
Liao, T., Wang, G., Yang, B., Lee, R., Pister, K. S. J., Levine, S., & Calandra, R. (2019) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2019.8793802 · Cited by 67
“robots with non-fixed or reconfigurable morphologies [11], [12], [13], [14], [15], where the controller must handle changes in the robot’s physical configuration, either to cope with unanticipated damage [13], [14] or prepare for different tasks [15].”
BackgroundCPG-Based Gait Generation of the Curved-Leg Hexapod Robot with Smooth Gait Transition
Bai, L., Hu, H., Chen, X., Sun, Y., Ma, C., & Zhong, Y. (2019) · Sensors · doi:10.3390/s19173705 · Cited by 51
Approximating gradients for differentiable quality diversity in reinforcement learning
Tjanaka, B., Fontaine, M. C., Togelius, J., & Nikolaidis, S. (2022) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3512290.3528705 · Cited by 76
“during deployment, the agent can adapt to the damage by selecting a policy that did not move the damaged foot during training [9, 13].”
BackgroundMethodInfluentialGaier, A., Asteroth, A., & Mouret, J.-B. (2017) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3071178.3071282 · Cited by 66
“Modern surrogate-assisted optimization often takes place within the framework of Bayesian optimization (BO) [2, 4, 23].”
BackgroundMethodInfluentialSelf-citationOn the Importance of Environments in Human-Robot Coordination
Fontaine, M. C., Hsu, Y.-C., Zhang, Y., Tjanaka, B., & Nikolaidis, S. (2021) · Robotics: Science and Systems · doi:10.15607/rss.2021.xvii.038 · Cited by 70
Nguyen, A., Yosinski, J., & Clune, J. (2016) · Evolutionary Computation · doi:10.1162/evco_a_00189 · Cited by 45
“…et al., 2015), and EAs can evolve neural networks to produce a variety of robot behaviors (Floreano and Mattiussi, 2008; Floreano and Keller, 2010; Cully et al., 2015; Clune et al., 2011; Li et al., 2014; Clune et al., 2009; Lee et al., 2013; Cheney et al., 2013), so we are optimistic that an…”
BackgroundMethodSelf-citationLearning Generalizable Locomotion Skills with Hierarchical Reinforcement Learning
Li, T., Lambert, N., Calandra, R., Meier, F., & Rai, A. (2020) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra40945.2020.9196642 · Cited by 54
“[12], [32], [33], [34] use a cyclic controller structure”
BackgroundJago, A. S., Raveendhran, R., Fast, N. J., & Gratch, J. S. (2023) · Journal of Experimental Social Psychology · doi:10.1016/j.jesp.2023.104553 · Cited by 54
“…modern algorithms are increasingly sophisticated in their ability to both process large amounts of data as well as adapt given new or missing data (Cully, Clune, Tarapore, & Mouret, 2015; James, Witten, Hastie, & Tibshirani, 2013), and algorithms are increasingly deployed in numerous human…”
BackgroundRecovery of locomotion after injury in Drosophila depends on proprioception
Isakov, A., Buchanan, S. M., Sullivan, B., Ramachandran, A., Chapman, J. K. S., Lu, E. S., Mahadevan, L., & de Bivort, B. (2016) · Journal of Experimental Biology · doi:10.1242/jeb.133652 · Cited by 47
“Moreover, the idea of plasticity after limb injury is not limited to the animal world: even in engineered systems such as legged robots, instilling the ability to recover locomotion after injury is an active research topic (Christensen et al., 2013; Cully et al., 2015).”
BackgroundMethodFunneled Bayesian Optimization for Design, Tuning and Control of Autonomous Systems
Martínez-Cantín, R. (2018) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2018.2805695 · Cited by 57
Stanley, K. O. (2019) · Artificial Life · doi:10.1162/artl_a_00294 · Cited by 93
“Repertoires of capabilities (as we are starting to see with quality diversity algorithms [10, 21]) could be generated in single runs, yielding not single solutions or controllers, but entire libraries of advanced behaviors.”
BackgroundTillmann, P., Jäger, K., & Becker, C. (2019) · Sustainable Energy & Fuels · doi:10.1039/c9se00750d · Cited by 48
Star-nose-inspired multi-mode sensor for anisotropic motion monitoring
Wang, Z., Bi, P., Yang, Y., Ma, H., Lan, Y., Sun, X., Hou, Y., Yu, H., Lu, G., Jiang, L., et al. (2020) · Nano Energy · doi:10.1016/j.nanoen.2020.105559 · Cited by 50
Virtual to Real-World Transfer Learning: A Systematic Review
Ranaweera, M., & Mahmoud, Q. H. (2021) · Electronics · doi:10.3390/electronics10121491 · Cited by 53
“[60] in their paper, proposed a novel algorithm called Intelligent Trial and Error to create a behavior–performance map to allow robots to rapidly adapt to their complex environments and recover from damages/injuries similarly to an animal.”
BackgroundAlgorithmic unconscious: why psychoanalysis helps in understanding AI
Possati, L. M. (2020) · Humanities and Social Sciences Communications · doi:10.1057/s41599-020-0445-0 · Cited by 53
“Cully et al. (2015) demonstrate that it is possible to construct 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-speci fi ed contingency plans ” (503).”
BackgroundBaines, R. L., Fish, F. E., Bongard, J., & Kramer-Bottiglio, R. K. (2024) · Nature Reviews Materials · doi:10.1038/s41578-024-00711-z · Cited by 54
“Across every conceivable performance metric — speed, efficiency, cognition, dexterity and mobility — biology reigns supreme 2–4 .”
Qualities, challenges and future of genetic algorithms: a literature reviewPreprint
Vié, A., Kleinnijenhuis, A. M., & Farmer, D. J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.05277 · Cited by 48
“Recently, related applications have conducted to improve the performance of robots behavior exposed to situations or damage not encountered before (Cully et al., 2015), enabling more robust, autonomous robots that evolve similarly to animals.”
Bayesian optimization with adaptive kernels for robot control
Martínez-Cantín, R. (2017) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2017.7989380 · Cited by 55
“Recent studies have found connections between Bayesian optimization and the way biological systems adapt and search in nature [3].”
BackgroundVachtsevanos, G., Lee, B., Oh, S., & Balchanos, M. G. (2018) · Journal of Intelligent & Robotic Systems · doi:10.1007/s10846-018-0881-x · Cited by 51
“Cully et al. suggest an improved trial and error method to determine the optimal action for a walking hexapod with a broken leg [39], but the large original search space and several minutes of lengthy adaptation time to the next step hamper their development.”
MethodFast and stable MAP-Elites in noisy domains using deep grids
Flageat, M., & Cully, A. (2020) · Conference on Artificial Life (ALIFE) · doi:10.1162/isal_a_00316 · Cited by 56
“…to generate an offspring, that will be added back to the cell it belongs to, randomly replacing one of the individuals contained in it. control (Cully et al., 2015; Cully and Mouret, 2015; Duarte et al., 2018); or in computer-aided design to generate multiple shapes of high-speed bicycles with…”
InfluentialSelf-citationDeep Reinforcement Learning Based Active Pantograph Control Strategy in High-Speed Railway
Wang, H., Han, Z., Liu, Z., & Wu, Y. (2022) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2022.3205452 · Cited by 46
“Since 2013, the deep reinforcement learning (DRL) technology has made considerable achievements and obtained impressive results in robot control [24], [25], unmanned aerial vehicle (UAV) control [26], [27], and other fields that are difficult to 0018-9545 © 2022 IEEE.”
BackgroundLearning a Contact-Adaptive Controller for Robust, Efficient Legged LocomotionPreprint
Da, X., Xie, Z., Hoeller, D., Boots, B., Anandkumar, A., Zhu, Y., Babich, B., & Garg, A. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2009.10019 · Cited by 80
“Many approaches like meta learning [15] or Bayesian Optimization in behavior space [16] have been proposed to adaptively update the policies.”
Worth of prior knowledge for enhancing deep learning
Xu, H. J., Chen, Y., & Zhang, D. (2024) · Nexus · doi:10.1016/j.ynexs.2024.100003 · Cited by 45
Wang, G., Zhang, C., Liu, S., Zhang, Y., Zhang, Y., & Wang, L. (2025) · Journal of Manufacturing Systems · doi:10.1016/j.jmsy.2025.06.014 · Cited by 50
Model-predictive control with stochastic collision avoidance using Bayesian policy optimization
Andersson, O., Wzorek, M., Rudol, P., & Doherty, P. (2016) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2016.7487661 · Cited by 54
“To demonstrate the approach we attempt to set up reasonably realistic scenarios involving the quadcopter model from section V and non-cooperative moving obstacles under uncertainty from section III.”
MethodMassively parallel and multiparameter titration of biochemical assays with droplet microfluidics
Baccouche, A., Okumura, S., Sieskind, R., Henry, E., Aubert-Kato, N., Bredèche, N., Bartolo, J.-F., Taly, V., Rondelez, Y., Fujii, T., et al. (2017) · Nature Protocols · doi:10.1038/nprot.2017.092 · Cited by 52
Designing Decisions in the Unknown: A Generative Model
Masson, P. L., Hatchuel, A., Glatin, M. L., & Weil, B. (2018) · European Management Review · doi:10.1111/emre.12289 · Cited by 50
Multi-objective quality diversity optimization
Pierrot, T., Richard, G., Beguir, K., & Cully, A. (2022) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3512290.3528823 · Cited by 65
“Introducing diversity in the descriptor space allows to retrieve solutions with lower quality (dominated by others) but potentially with better properties, such as robustness, a property already observed for map-elites [8].”
BackgroundInfluentialSelf-citationIntelligence-Augmented Rat Cyborgs in Maze Solving
Yu, Y., Pan, G., Gong, Y., Xu, K., Zheng, N., Hua, W., Zheng, X., & Wu, Z. (2016) · PLoS ONE · doi:10.1371/journal.pone.0147754 · Cited by 46
“Moreover, the rat cyborgs are agile in different types of terrain (see S1 Video), and have the potential to solve unanticipated problems relying on instinct [38].”
BackgroundThe evolution of two types of play
Smaldino, P. E., Palagi, E., Burghardt, G. M., & Pellis, S. M. (2019) · Behavioral Ecology · doi:10.1093/beheco/arz090 · Cited by 53
A Fast Online Frequency Adaptation Mechanism for CPG-Based Robot Motion Control
Thor, M., & Manoonpong, P. (2019) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2019.2926660 · Cited by 49
“[5] showed how a repertoire of locomotion”
A 3D Printed Paper-Based Thermally Driven Soft Robotic Gripper Inspired by Cabbage
Hu, F., Lyu, L., & He, Y. (2019) · International Journal of Precision Engineering and Manufacturing · doi:10.1007/s12541-019-00199-6 · Cited by 46
Cully, A. (2021) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3449639.3459326 · Cited by 57
“r;y gripper), while the fitness is the opposite of variance of the joint’s articular position:fit(x) = Var(x) Hexapod-uni and Hexapod-omni These two tasks are based on a simulated hexapod introduced in Cully et al. (2015) and reused in multiple works in the literature (Cully and Demiris, 2018; Vassiliades and Mouret, 20…”
BackgroundInfluentialSelf-citationChallenges, evaluation and opportunities for open-world learning
Kejriwal, M., Kildebeck, E., Steininger, R. J., & Shrivastava, A. (2024) · Nature Machine Intelligence · doi:10.1038/s42256-024-00852-4 · Cited by 62
“Rather than relying on slow re-training, OWL algorithms, especially in physically embodied applications like robotics, need to be resilient and ‘adapt like animals’ 16 .”
Elastic MXene Hydrogel Microfiber-Derived Electronic Skin for Joint Monitoring
Guo, J., Yu, Y., Zhang, H., Sun, L., & Zhao, Y. (2021) · ACS Applied Materials & Interfaces · doi:10.1021/acsami.1c10311 · Cited by 43
Hamann, H., Khaluf, Y., Botev, J., Soorati, M. D., Ferrante, E., Kosak, O., Montanier, J.-M., Mostaghim, S., Redpath, R., Timmis, J., et al. (2016) · Frontiers in Robotics and AI · doi:10.3389/frobt.2016.00014 · Cited by 49
“Solutions range from the restriction of the search space (Koos et al., 2013; Cully et al., 2015) to the design of behaviors during the deployment of the system (Watson et al., 2002; Bredeche and Montanier, 2010). he design of a hybrid-society system could beneit from these approaches.”
BackgroundAn origami-wheeled robot with variable width and enhanced sand walking versatility
Liu, J., Pang, Z., Li, Z., Wen, G., Su, Z., He, J., Liu, K., Jiang, D., Liu, K., Chen, S., et al. (2024) · Thin-Walled Structures · doi:10.1016/j.tws.2024.112645 · Cited by 42
Snapbot: A reconfigurable legged robot
Kim, J., Alspach, A. N., & Yamane, K. (2017) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros.2017.8206477 · Cited by 48
“More recently, researchers have used machine learning techniques to developed robots that can adapt like animals when some part of the body is damaged [14].”
BackgroundMAP-Elites Enables Powerful Stepping Stones and Diversity for Modular Robotics
Nordmoen, J., Veenstra, F., Ellefsen, K. O., & Glette, K. (2021) · Frontiers in Robotics and AI · doi:10.3389/frobt.2021.639173 · Cited by 54
“This corroborates the previous findings mentioned in (Cully and Demiris, 2017) of curiosity being a useful addition to MAP-Elites.”
BackgroundMethodResultInfluentialBioinspired Ultrathin Piecewise Controllable Soft Robots
Li, D., Wang, S., He, J., Zeng, H., Yao, K., Gao, Z., Wu, M., Liu, Y., Wang, L., Xie, Z., et al. (2021) · Advanced Materials Technologies · doi:10.1002/admt.202001095 · Cited by 43
Bayesian Optimization with Automatic Prior Selection for Data-Efficient Direct Policy Search
Pautrat, R., Chatzilygeroudis, K. I., & Mouret, J.-B. (2018) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2018.8463197 · Cited by 50
Self-citationHierarchical behavioral repertoires with unsupervised descriptors
Cully, A., & Demiris, Y. (2018) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3205455.3205571 · Cited by 49
“enerated and organized automatically, to form a continuous behavioral space, that can be useful to learn other behaviors (e.g., dance moves) or adapt to unforeseen situations (e.g., damage situations [3]). One challenge with HBRs is that chaining behavior executions make them start from arbitrary initial conditions. „…”
BackgroundMethodInfluentialSelf-citationSurface Actuation and Sensing of a Tensegrity Structure Using Robotic Skins
Booth, J., Cyr-Choinière, O., Case, J. C., Shah, D., Yuen, M. C., & Kramer-Bottiglio, R. K. (2020) · Soft Robotics · doi:10.1089/soro.2019.0142 · Cited by 46
“While this approach has the disadvantage of needing to be trained on every individual robot, it offers flexibility if the robot is damaged, since the model-free approach can be retrained during a mission.(47) When all actuators are at rest, the tensegrity structure is stretched more toward the top than the bottom of t…”
Illuminating diverse neural cellular automata for level generation
Earle, S., Snider, J., Fontaine, M. C., Nikolaidis, S., & Togelius, J. (2022) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3512290.3528754 · Cited by 41
“We train CPPNs via MAP-Elites with the variation operations from the NEAT algorithm [38].”
MethodInfluentialAdvancement of machine learning in materials science
Rajendra, P., Girisha, A., & Naidu, T. G. (2022) · Materials Today Proceedings · doi:10.1016/j.matpr.2022.04.238 · Cited by 45
Fault-Tolerant Tripod Gait Planning and Verification of a Hexapod Robot
Liu, Y., Fan, X., Ding, L., Wang, J., Liu, T., & Gao, H. (2020) · Applied Sciences · doi:10.3390/app10082959 · Cited by 40
“Antoine Cully created a new technology to create a detailed high-performance behavior space map using the Gauss process model of machine learning, and an intelligent trial-and-error algorithm was introduced that can allow the robot to adapt to the environment successfully, including with an injured, broken, or missing…”
MethodReality-Assisted Evolution of Soft Robots through Large-Scale Physical Experimentation: A Review
Howison, T., Hauser, S., Hughes, J., & Iida, F. (2020) · Artificial Life · doi:10.1162/artl_a_00330 · Cited by 43
“2015 [19] Hexapod C ODE IT&E 10 10 Rosendo et al.”
BackgroundMethodInfluentialArtificial intelligence interventions in 2D MXenes-based photocatalytic applications
Mahapatra, D. M., Kumar, A., Kumar, R., Gupta, N. K., Ethiraj, B., & Singh, L. (2025) · Coordination Chemistry Reviews · doi:10.1016/j.ccr.2025.216460 · Cited by 39
Diverse Agents for Ad-Hoc Cooperation in Hanabi
Canaan, R., Togelius, J., Nealen, A., & Menzel, S. (2019) · IEEE Conference on Games (CoG) · doi:10.1109/cig.2019.8847944 · Cited by 51
“MAP-Elites was first proposed to pre-compute a variety of effective gaits for a sixlegged robot, so that, when the robot suffers damage, it can quickly search for a gait that adapts to the damage and allows it to keep moving at a decent pace [25].”
BackgroundFault Detection in a Swarm of Physical Robots Based on Behavioral Outlier Detection
Tarapore, D., Timmis, J., & Christensen, A. L. (2019) · IEEE Transactions on Robotics · doi:10.1109/tro.2019.2929015 · Cited by 39
“Accommodation procedures may comprise caging the faulty robot, diagnosing the nature of the fault [24], and consequently adapting the behavior of the faulty robot [25].”
BackgroundSelf-citationUnsupervised Learning and Exploration of Reachable Outcome Space
Paolo, G., Laflaquière, A., Coninx, A., & Doncieux, S. (2020) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra40945.2020.9196819 · Cited by 52
“An AE, combined with MAP-Elites [2], was also used in [29] to generate and classify novel images.”
BackgroundMethodInfluentialGraule, M. A., McCarthy, T. P., Teeple, C. B., Werfel, J., & Wood, R. J. (2022) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2022.3149580 · Cited by 49
Flexible, lightweight, tunable robotic arms enabled by X-tensegrity inspired structures
Yue, X.-H., Xu, Y., Sun, Z.-Y., Liu, L.-Y., Wang, Y., Xu, G., Cao, C. C., & Zhang, L. (2024) · Composite Structures · doi:10.1016/j.compstruct.2024.118331 · Cited by 38
“These devices encompass various technological domains including mechanical control, power electronics, and artificial intelligence [1 – 3].”
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.