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; Learning Robust Task Priorities and Gains for Control of Redundant Robots 2. the name, complete physical mailing address, e-mail address, and phone number of EACH author of EACH paper(s); Luigi Penco Inria Nancy - Grand Est, 615 rue du Jardin Botanique, 54600 Villers-lès-Nancy, France luigi.penco@inria.fr (+33) 3 83 59 30 00 Enrico Hoffman Istituto Italiano di Tecnologia, Via Morego, 30, 16163 Genova GE, Italy enrico.mingo@iit.it (+39) 010 71781 Valerio Modugno Sapienza University of Rome, Via Ariosto 25, 00185, Rome - Italy modugno@diag.uniroma1.it (+39)06-77274-158 Waldez Azevedo-Gomes Junior Inria Nancy - Grand Est, 615 rue du Jardin Botanique, 54600 Villers-lès-Nancy, France waldez.azevedo-gomes-junior@inria.fr (+33) 3 83 59 30 00 Jean-Baptiste Mouret Inria Nancy - Grand Est, 615 rue du Jardin Botanique, 54600 Villers-lès-Nancy, France jean-baptiste.mouret@inria.fr (+33) 3 83 59 30 00 Serena Ivaldi Inria Nancy - Grand Est, 615 rue du Jardin Botanique, 54600 Villers-lès-Nancy, France serena.ivaldi@inria.fr (+33) 3 83 59 30 00 3. the name of the corresponding author (i.e., the author to whom notices will be sent concerning the competition); Jean-Baptiste Mouret jean-baptiste.mouret@inria.fr 4. the abstract of the paper(s); Generating complex movements in redundant robots like humanoids is usually done by means of multi-task controllers based on quadratic programming, where a multitude of tasks is organized according to strict or soft priorities. Time-consuming tuning and expertise are required to choose suitable task priorities, and to optimize their gains. Here, we automatically learn the controller configuration (soft and strict task priorities and Convergence Gains), looking for solutions that track a variety of desired task trajectories efficiently while preserving the robot's balance. We use multi-objective optimization to compare and choose among Pareto-optimal solutions that represent a trade-off of performance and robustness and can be transferred onto the real robot. We experimentally validate our method by learning a control configuration for the iCub humanoid, to perform different whole-body tasks, such as picking up objects, reaching and opening doors. Video (including results with the robot): https://www.youtube.com/watch?v=MokoPcHvhlQ 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. (G) The result solves a problem of indisputable difficulty in its field. Humanoid robots have been a long standing dream for robotics because they can be as versatile as humans, but they need to synchronize dozen (often 30+) motors precisely to keep their balance while executing complex whole-body motions, such as lifting objects from the ground. After decades of research, the community converged on controllers that are based on online quadratic optimization: at each time-step the robot searches for the minimum of a cost function that describes how well the main parts (e.g., the hands) follow the desired trajectory while taking into account the constraints of the robot (kinematics, keeping the center of mass in the support polygon, etc.). Nevertheless, this optimization-based formulation introduced dozen of parameters to tune (should the robot prioritize the hand or the center of mass? and what about the feet?). In a typical humanoid controller, there are about 10-15 "tasks" (objectives), each of them with a priority and a gain (i.e., at least 30 parameters). Importantly, the priority is not simply a weight, but a change in the structure of a hierarchical quadratic program. All the humanoid robots controllers have so-far been manually tuned by expert engineers. This is a time-consuming process that most of the time leads to sub-optimal results. In addition, these whole-body controllers are usually tuned for specific trajectories and they therefore need to be changed for each "demonstration". In this work, we used NSGA-II to optimize the priorities and the gains of a whole-body controller. The results are used on the state-of-the-art iCub humanoid robot (Video: https://www.youtube.com/watch?v=MokoPcHvhlQ) that is teleoperated in real-time using a motion capture suit. Two objectives are optimized in simulation: how well the robot tracks the main parts of the operator (mainly the hands) and their posture in complex trajectories, and how stable the robot is. There is a clear trade-off because following the desired trajectory perfectly leads to "agressive behaviors" that make the robot falls. Having a Pareto front instead of a single optimized solution makes it possible to select which trade-off works best on the robot. In particular, we observed experimentally that the solutions that we the most high-performing in simulation makes the robot fall (a typical instance of the "reality gap"). Our iCub robot now uses one of the evolutionary optimized controller, which outperforms the hand-designed controller in all the scenarios that we tested. 7. a full citation of the paper (that is, author names; title, publication date; name of journal, conference, or book in which article appeared; name of editors, if applicable, of the journal or edited book; publisher name; publisher city; page numbers, if applicable); Penco L, Hoffman EM, Modugno V, Gomes W, Mouret JB, Ivaldi S. Learning robust task priorities and gains for control of redundant robots. IEEE Robotics and Automation Letters. 2020 Feb 10;5(2):2626-33. 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 equally among the co-authors. 9. a statement stating why the authors expect that their entry would be the "best," and The evolutionary computation community has been trying to propose ideas to the robotics community since their inception (see the evolutionary robotics / Complex system track at GECCO). In our opinion, this is one of the most convincing use of evolutionary computation for state-of-the-art robotics, and we received requests to test our code from several humanoid robotics teams. Most teams are impressed that the results are demonstrated on a real humanoid robot, not in simulation, and that is compatible with their approaches. In our team, we just applied the same technique to tune the controller of our new robot (a full-size humanoid): https://youtu.be/cnIo-aWCOcs?t=39 In addition, "pushing" Pareto-based multi-objective optimization to robotics is very inspiring for the robotics community because it is only used to weight-based aggregation of objectives. 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), GI (genetic improvement), GE (grammatical evolution), GEP (gene expression programming), DE (differential evolution), etc. Multi-objective evolutionary algorithms. 11. The date of publication of each paper. If the date of publication is not on or before the deadline for submission, but instead, the paper has been unconditionally accepted for publication and is “in press” by the deadline for this competition, the entry must include a copy of the documentation establishing that the paper meets the "in press" requirement. February 2020. 12. Link to the paper (open-access): https://hal.inria.fr/hal-02456663/document