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 .”
Diagnosis and Fault-Tolerant Control
(2006) · Springer eBooks · doi:10.1007/978-3-540-35653-0 · Cited by 1,228
Postural Hand Synergies for Tool Use
Santello, M., Flanders, M., & Soechting, J. F. (1998) · The Journal of Neuroscience · doi:10.1523/jneurosci.18-23-10105.1998 · Cited by 1,215 · cited 1×
“For grasping, performance could be the amount of surface contact, and it has been demonstrated that 90% of effective poses for the 21-degree-of-freedom human hand can be captured by a 3-dimensional behavioral space 20 .”
BackgroundWagner, U., Gais, S., Haider, H., Verleger, R., & Born, J. (2004) · Nature · doi:10.1038/nature02223 · Cited by 1,012 · cited 1×
“This process is reminiscent of the finding that some animals start the day with new ideas that they may quickly disregard after experimenting with them 27 , and more generally, that sleep improves creativity on cognitive tasks 28 .”
BackgroundAcceptance of Healthcare Robots for the Older Population: Review and Future Directions
Broadbent, E., Stafford, R., & MacDonald, B. (2009) · International Journal of Social Robotics · doi:10.1007/s12369-009-0030-6 · Cited by 887 · cited 1×
“As described in experiment 5 of the Supplementary Information, the evaluated behaviour descriptors characterize the following: (1) time each leg is in contact with the ground (‘Duty factor’); (2) orientation of the robot frame (‘Orientation’); (3) instantaneous velocity of the robot (‘Displacement’); (4) energy expend…”
BackgroundThe motor infrastructure: from ion channels to neuronal networks
Grillner, S. (2003) · Nature Reviews Neuroscience · doi:10.1038/nrn1137 · Cited by 839
Resilient Machines Through Continuous Self-Modeling
Bongard, J., Zykov, V., & Lipson, H. (2006) · Science · doi:10.1126/science.1133687 · Cited by 709
Perspectives and problems in motor learning
Wolpert, D. M., Ghahramani, Z., & Flanagan, J. R. (2001) · Trends in Cognitive Sciences · doi:10.1016/s1364-6613(00)01773-3 · Cited by 667
Probabilistic brains: knowns and unknowns
Pouget, A., Beck, J. M., Ma, W. J., & Latham, P. E. (2013) · Nature Neuroscience · doi:10.1038/nn.3495 · Cited by 615 · 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 .”
MethodNagatani, K., Kiribayashi, S., Okada, Y., Otake, K., Yoshida, K., Tadokoro, S., Nishimura, T., Yoshida, T., Koyanagi, E., Fukushima, M., et al. (2013) · Journal of Field Robotics · doi:10.1002/rob.21439 · Cited by 602 · cited 1×
“As described in experiment 5 of the Supplementary Information, the evaluated behaviour descriptors characterize the following: (1) time each leg is in contact with the ground (‘Duty factor’); (2) orientation of the robot frame (‘Orientation’); (3) instantaneous velocity of the robot (‘Displacement’); (4) energy expend…”
BackgroundPolicy gradient reinforcement learning for fast quadrupedal locomotion
Kohl, N., & Stone, P. (2004) · IEEE International Conference on Robotics and Automation, 2004. Proceedings. ICRA '04. 2004 · doi:10.1109/robot.2004.1307456 · Cited by 564
Bayesian Approach to Global Optimization--Theory and Applications.
Boender, C. G. E., & Mockus, J. (1991) · Mathematics of Computation · doi:10.2307/2008419 · Cited by 409 · cited 3×
“It is called Bayesian because, in its general formulation 22 , this algorithm chooses the next point by computing a posterior distribution of the objective function using the likelihood of the data already acquired and a prior on the type of function.”
BackgroundMethodIlluminating search spaces by mapping elites
Mouret, J.-B., & Clune, J. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1504.04909 · Cited by 386 · cited 1×
“3c, the undamaged robot takes 24 s [16 s; 41 s] in 3 [2; 5] physical trials and the repaired robot takes 29 s [16 s; 82 s] in 3.”
BackgroundSelf-citationActivities of the rescue robots at the World Trade Center from 11-21 september 2001
Murphy, R. R. (2004) · IEEE Robotics & Automation Magazine · doi:10.1109/mra.2004.1337826 · Cited by 325
How sleep affects the developmental learning of bird song
Derégnaucourt, S., Mitra, P. P., Fehér, O., Pytte, C., & Tchernichovski, O. (2005) · Nature · doi:10.1038/nature03275 · Cited by 320 · cited 1×
“This process is reminiscent of the finding that some animals start the day with new ideas that they may quickly disregard after experimenting with them 27 , and more generally, that sleep improves creativity on cognitive tasks 28 .”
BackgroundInnovative problem solving by wild spotted hyenas
Benson-Amram, S., & Holekamp, K. E. (2012) · Proceedings of the Royal Society B: Biological Sciences · doi:10.1098/rspb.2012.1450 · Cited by 287 · cited 2×
“Also like animals 24 , Intelligent Trial and Error allows the quick identification of working behaviors with a few, diverse tests instead of trying behaviors at random or trying small modifications to the best behavior found so far.”
BackgroundThe Bayesian approach to global optimization
Mockus, J. (2005) · Lecture Notes in Control and Information Sciences · doi:10.1007/bfb0006170 · Cited by 283
How UGVs physically fail in the field
Carlson, J., & Murphy, R. R. (2005) · IEEE Transactions on Robotics · doi:10.1109/tro.2004.838027 · Cited by 272 · cited 2×
“3C, undam-aged: 24 [16; 41] seconds, 3 [2; 5] physical trials, repaired: 29 [16; 82] seconds, 3.5 [2; 10] trials).”
BackgroundInfluentialAutomatic gait optimization with Gaussian process regression
Lizotte, D. J., Wang, T., Bowling, M., & Schuurmans, D. (2007) · Cited by 237
Verma, V., Gordon, G., Simmons, R., & Thrun, S. (2004) · IEEE Robotics & Automation Magazine · doi:10.1109/mra.2004.1310942 · Cited by 219
Learning to Move in Modular Robots using Central Pattern Generators and Online Optimization
Badri–Spröwitz, A., Möckel, R., Maye, J., & Ijspeert, A. J. (2008) · The International Journal of Robotics Research · doi:10.1177/0278364907088401 · Cited by 144
Antonelli, G., Fossen, T. I., & Yoerger, D. (2008) · Springer eBooks · doi:10.1007/978-3-540-30301-5_44 · Cited by 89 · cited 2×
“3C, undam-aged: 24 [16; 41] seconds, 3 [2; 5] physical trials, repaired: 29 [16; 82] seconds, 3.5 [2; 10] trials).”
BackgroundInfluentialUsing response surfaces and expected improvement to optimize snake robot gait parameters
Tesch, M., Schneider, J., & Choset, H. (2011) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros.2011.6095076 · Cited by 86
An experimental comparison of Bayesian optimization for bipedal locomotion
Calandra, R., Andre Seyfarth, A., Peters, J., & Deisenroth, M. P. (2014) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2014.6907117 · Cited by 78 · cited 1×
“For the five damage scenarios, the robot adapts in approximately one minute (66 s [24 s; 134 s] in 8 [3; 16] trials).”
BackgroundChristensen, D. J., Schultz, U. P., & Støy, K. (2013) · Robotics and Autonomous Systems · doi:10.1016/j.robot.2013.05.009 · Cited by 64
Kinematic and kinetic analysis of dogs during trotting after amputation of a pelvic limb
Hogy, S. M., Worley, D. R., Jarvis, S. L., Hill, A. E., Reiser, R. F., & Haussler, K. K. (2013) · American Journal of Veterinary Research · doi:10.2460/ajvr.74.9.1164 · Cited by 63 · cited 1×
“Injured animals respond differently: they learn by trial and error how to compensate for damage (e.g. learning which limp minimizes pain) 15,16 .”
BackgroundKinematic and kinetic analysis of dogs during trotting after amputation of a thoracic limb
Jarvis, S. L., Worley, D. R., Hogy, S. M., Hill, A. E., Haussler, K. K., & Reiser, R. F. (2013) · American Journal of Veterinary Research · doi:10.2460/ajvr.74.9.1155 · Cited by 61
Bayesian optimization explains human active search
Borji, A., & Itti, L. (2013) · Cited by 53
Ground reaction force adaptations to tripedal locomotion in dogs
Fuchs, A., Goldner, B., Nolte, I. J. A., & Schilling, N. (2014) · The Veterinary Journal · doi:10.1016/j.tvjl.2014.05.012 · Cited by 35 · cited 3×
“3C, undam-aged: 24 [16; 41] seconds, 3 [2; 5] physical trials, repaired: 29 [16; 82] seconds, 3.5 [2; 10] trials).”
BackgroundInfluentialSanderson, K. (2010) · Nature · doi:10.1038/463600a · Cited by 22 · cited 2×
“A major obstacle to their widespread adoption in more complex environments outside of factories is their fragility 4,5 : Robots presently pale in comparison to natural animals in their ability to invent compensatory behaviors after an injury (Fig.”
Background
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
Deep neural networks are easily fooled: High confidence predictions for unrecognizable images
Nguyen, A., Yosinski, J., & Clune, J. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298640 · Cited by 3,258
Self-citationDomain randomization for transferring deep neural networks from simulation to the real world
Tobin, J., Fong, R., Ray, A., Schneider, J., Zaremba, W., & Abbeel, P. (2017) · 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros.2017.8202133 · Cited by 3,124
“[8] demonstrates that exploration in the real world can be more efficient with prior knowledge from a simulator.”
BackgroundAdversarial Examples: Attacks and Defenses for Deep Learning
Yuan, X., He, P., Zhu, Q., & Li, X. (2019) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2018.2886017 · Cited by 1,733
“Then in each iteration, MAP-Elites, like general EA algorithm, chose a random organism, mutated them randomly, and replaced with the current ones if the new ones have higher fitness (high certainty for a class of a neural network).”
MethodInfluentialRahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J.-F., Breazeal, C., Crandall, J. W., Christakis, N. A., Couzin, I. D., Jackson, M. O., et al. (2019) · Nature · doi:10.1038/s41586-019-1138-y · Cited by 1,364
Machine learning in materials science
Wei, J., Chu, X., Sun, X., Xu, K., Deng, H., Chen, J., Wei, Z., & Lei, M. (2019) · InfoMat · doi:10.1002/inf2.12028 · Cited by 1,069
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.