1. The complete title of one (or more) paper(s) published in the open literature describing the work that the author claims describes a human-competitive result; Robots that can adapt like animals 2. The name, complete physical mailing address, e-mail address, and phone number of EACH author of EACH paper(s); Antoine Cully a.cully@imperial.ac.uk ISN Group, Level 10, Room 1006 Electrical and Electronic Engineering Building Imperial College London SW7 2BT, UK Phone: +44 (0)207 594 6316 Jeff Clune jeffclune@uwyo.edu Computer Science University of Wyoming Laramie, WY 82070, USA phone: 517 214-1060 (mobile) Danesh Tarapore danesh.tarapore@york.ac.uk Electronics University of York Heslington York, YO10 5DD United Kingdom Phone: +44 (0)74 814 93211 Jean-Baptiste Mouret jean-baptiste.mouret@inria.fr Inria Nancy - Grand Est 615 Rue du Jardin botanique 54600 Villers-lC(s-Nancy, France phone: + 33 (0) 3 54 95 86 16 3. The name of the corresponding author (i.e., the author to whom notices will be sent concerning the competition); Jean-Baptiste Mouret 4. The abstract of the paper(s); 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 bthink outside the boxb 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 robotbs 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. VIDEO: https://www.youtube.com/watch?v=T-c17RKh3uE 5. A list containing one or more of the eight letters (A, B, C, D, E, F, G, or H) that correspond to the criteria (see above) that the author claims that the work satisfies; E, G 6. A statement stating why the result satisfies the criteria that the contestant claims (see examples of statements of human-competitiveness as a guide to aid in constructing this part of the submission); (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. Making robots that can adapt to damage is a long-standing challenge of human engineering. To date, most efforts have focused on manually designed self-diagnosis algorithms and robust control, that is, on human-based engineering; however, such approaches require anticipating all the ways that things can go wrong, and thus do not scale to realistic situations. Our new approach, which was featured on the cover of Nature, combines evolutionary algorithms with Bayesian optimization to rapidly learn on the fly on a real robot how to adapt to any damage that occurs to that robot. Importantly, and unprecedentedly in the history of robotics, such adaptation takes only 1-2 minutes. Specifically, our technique leverages a novel evolutionary algorithm (MAP-Elites) to create a low-dimensional, diverse search space that, in turn, allows the robot to quickly adapt to unexpected damage conditions. Automating the creation of the search space for online adaptation via an evolutionary algorithm is the key innovation that makes our method competitive with human-based engineering: without this step, trial-and-error learning algorithms require hundreds of trials, that is, hours of learning time, which makes them non-competitive with the more traditional approaches; in contrast, our method allows robots to find a compensatory behaviors after only a dozen of trials (we demonstrated it on a 6-legged robot and a robotic arm, where each trials lasted 5 seconds). (G) The result solves a problem of indisputable difficulty in its field. The recent DARPA robotics challenge highlighted that robots are ill equipped to deal with unexpected situations: most robots fell or failed in some other ways during the competition, despite the fact that they were remote-controlled and that the tasks are simple for humans (e.g. opening a door). This video shows a few failures: https://www.youtube.com/watch?v=g0TaYhjpOfo . In real situations, most deployed robots become damaged, preventing them from successfully performing their mission. For instance, a robot was sent to the World Trade Center after 9/11; the first failure appeared on average after 1.4 minutes (see "Carlson J, Murphy RR. How UGVs physically fail in the field. IEEE Transactions on Robotics,. 2005 Jun;21(3):423-37). More recently, several robots are still trapped in the Fukushima power plants because they were unable to recover from an unexpected situation. These failures demonstrate that autonomous damage recovery is a problem of indisputable difficulty and importance in robotics. The importance of our contribution for the field was recognized by Nature, which chose not only to publish the paper, but also to make it its cover story. The cover and paper can be viewed here: http://pages.isir.upmc.fr/~mouret/website/pics/nature_cover_cully.pdf http://www.isir.upmc.fr/files/2015ACLI3468.pdf [arxiv version] 7. A full citation of the paper (that is, author names; publication date; name of journal, conference, technical report, thesis, book, or book chapter; name of editors, if applicable, of the journal or edited book; publisher name; publisher city; page numbers, if applicable); Cully A, Clune J, Tarapore D, Mouret JB. Robots that can adapt like animals. Nature. 2015 May 28;521(7553):503-7. (cover story). 8. A statement either that "any prize money, if any, is to be divided equally among the co-authors" OR a specific percentage breakdown as to how the prize money, if any, is to be divided among the co-authors; Any prize money, if any, is to be divided as follows: Antoine Cully: 40% Jeff Clune: 20% Danesh Tarapore: 20% Jean-Baptiste Mouret: 20% 9. A statement stating why the authors expect that their entry would be the "best," and Evolutionary computation and robotics are two fields that share many ambitions, starting with the ambition to reproduce the marvels of nature in machines. However, despite the huge amount of work in evolutionary robotics, evolutionary computation has mostly failed to convince the robotics community that evolution could help solve the big problems in robotics. Our contribution is one of the very few evolutionary robotics articles that is deemed as very relevant to improve the state of the art in robotics. More generally, there is great skepticism in machine learning communities about evolutionary algorithms. Our paper was the cover story in Naturebs special issue about artificial intelligence, which also included an article summarizing Deep Learning (by Lecun, Bengio, and Hinton), Bayesian approaches to Probabilistic Reasoning (by Ghahramani), traditional Reinforcement Learning (by Littman), etc. That the cover story of that issue featured an Evolutionary Algorithm is an important milestone for our field and highlights the relevance of our contribution not only to Evolutionary Algorithms, but more broadly to artificial intelligence and robotics. 10. An indication of the general type of genetic or evolutionary computation used, such as GA (genetic algorithms), GP (genetic programming), ES (evolution strategies), EP (evolutionary programming), LCS (learning classifier systems), GE (grammatical evolution), GEP (gene expression programming), DE (differential evolution), etc. The algorithm is a new type of evolutionary algorithm that is called a "quality diversity algorithm" [1] (or "illumination algorithm" [2]). [1] Pugh, J. K., Soros, L. B., Szerlip, P. A., & Stanley, K. O. (2015, July). Confronting the challenge of quality diversity. In Proceedings of the 2015 on Genetic and Evolutionary Computation Conference (pp. 967-974). ACM. [2] Mouret, J. B., & Clune, J. (2015). Illuminating search spaces by mapping elites. arXiv preprint arXiv:1504.04909.