1. Title of the Publication: MorphVAE: Advancing Morphological Design of Voxel-Based Soft Robots with Variational Autoencoders 2. Author Information Junru Song Renmin University of China, No.59 Zhongguancun Street, Haidian District, Beijing, 100872, P.R. China woodysjr@foxmail.com (86)13080751013 Yang Yang Chinese Academy of Military Science, No.53 Dongdajie Street, Fengtai District, Beijing, 100071, P.R. China bigyangy@gmail.com (86)18514592959 Wei Peng Chinese Academy of Military Science, No.53 Dongdajie Street, Fengtai District, Beijing, 100071, P.R. China weipeng0098@126.com (86)18515012251 Weien Zhou Chinese Academy of Military Science, No.53 Dongdajie Street, Fengtai District, Beijing, 100071, P.R. China weienzhou@outlook.com (86)13261185761 Feifei Wang Renmin University of China, No.59 Zhongguancun Street, Haidian District, Beijing, 100872, P.R. China feifei.wang@ruc.edu.cn (86)13811190585 Wen Yao Chinese Academy of Military Science, No.53 Dongdajie Street, Fengtai District, Beijing, 100071, P.R. China wendy0782@126.com (86)18518169621 3. Corresponding Author Feifei Wang feifei.wang@ruc.edu.cn 4. Paper Abstract Soft robot design is an intricate field with unique challenges due to its complex and vast search space. In the past literature, evolutionary computation algorithms have shown potential in this realm. However, these methods are sample inefficient and predominantly focus on rigid robots in locomotion tasks, which limit their performance and application in robot design automation. In this work, we propose MorphVAE, an innovative probabilistic generative model that incorporates a multi-task training scheme and a meticulously crafted sampling technique termed "continuous natural selection", aimed at bolstering sample efficiency. This method empowers us to gain insights from assessed samples across diverse tasks and temporal evolutionary stages, while simultaneously maintaining a delicate balance between optimization efficiency and biodiversity. Through extensive experiments in various locomotion and manipulation tasks, we substantiate the efficiency of MorphVAE in generating high-performing and diverse designs, surpassing the performance of competitive baselines. 5. Competition Criterion (B) The result is equal to or better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal; (D) The result is publishable in its own right as a new scientific result independent of the fact that the result was mechanically created; (F) The result is equal to or better than a result that was considered an achievement in its field at the time it was first discovered. 6. Statement Why the Result Satisfies Criterion (B), (D) and (F) Reason for (B) and (F): Our method, termed MorphVAE, evolves the morphological design of voxel-based soft robots (VSRs) with an innovative combination of the population-based evolutionary algorithm and the variational autoencoder, a type of probabilistic generative model. We demonstrated through extensive experiments that our method outperforms a wide range of existing methods, including search strategies based on generative models, evolutionary algorithms, and EAs assisted with surrogate models, in terms of both optimization efficiency and bio-diversity. We list these baselines below, all of which were considered important achievements during their time of publication: (1) Genetic Algorithm (GA) - Zbigniew, M. (1996). Genetic algorithms+ data structures= evolution programs. Computational statistics, 24, 372-373; (2) Bayesian Optimization (BO) - Kushner, H. J. (1964). A new method of locating the maximum point of an arbitrary multipeak curve in the presence of noise; (3) Speciated Evolver (SE) - Medvet, E., Bartoli, A., Pigozzi, F., & Rochelli, M. (2021, June). Biodiversity in evolved voxel-based soft robots. In Proceedings of the Genetic and Evolutionary Computation Conference (pp. 129-137); (4) RoboGAN - Hu, J., Whitman, J., Travers, M., & Choset, H. (2022, May). Modular robot design optimization with generative adversarial networks. In 2022 International Conference on Robotics and Automation (ICRA) (pp. 4282-4288). IEEE; (5) Action Inheritance-Based Evolutionary Algorithm (AIEA) - Liu, S., Yao, W., Wang, H., Peng, W., & Yang, Y. (2023). Rapidly Evolving Soft Robots via Action Inheritance. IEEE Transactions on Evolutionary Computation; (6) Random Forest-Based Evolutionary Algorithm (RFEA) - Sun, Y., Wang, H., Xue, B., Jin, Y., Yen, G. G., & Zhang, M. (2019). Surrogate-assisted evolutionary deep learning using an end-to-end random forest-based performance predictor. IEEE Transactions on Evolutionary Computation, 24(2), 350-364. Reason for (D): Our work was published in the 38th AAAI Conference on Artificial Intelligence. The Association for the Advancement of Artificial Intelligence (AAAI) is the premier scientific society dedicated to advancing the scientific understanding of the mechanisms underlying thought and intelligent behavior and their embodiment in machines. Thus, accepted papers must meet high standards of scientific rigour, and they typically describe an exceptional contribution that surpasses existing technology. Our pioneering application of probabilistic generative models to evolutionary computation has also given rise to a patent application, confirming that our work is of significant commercial value as well. 7. Full Citation Song, J., Yang, Y., Peng, W., Zhou, W., Wang, F., & Yao, W. (2024, March). MorphVAE: Advancing Morphological Design of Voxel-Based Soft Robots with Variational Autoencoders. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 38, No. 9, pp. 10368-10376). 8. Prize Breakdown Statement Any prize money, if any, is to be divided equally among the co-authors. 9. Required Statement Indicating Why this Entry Could be the "Best" Our work addresses one of the major challenges in the field of Robotics, i.e. robot design automation, and presents a novel approach to morphological representation and optimization. By incorporating deep probabilistic generative models into the classic population-based evolutionary algorithms and meticulously designing a training procedure for this innovative combination, we pioneered a transformative paradigm within evolutionary computation which not only has been proven to significantly enhance optimization efficiency and bio-diversity in robot design automation, but also promises broad applicability across diverse domains, including but not limited to computational biology and material science. The novelty, state-of-the-art performance and versatile adaptability unequivocally position our work as a winning entry. 10. Evolutionary Computation Type ES (evolution strategies) 11. Publication Date 2024-03-24