Abstract
Reinforcement learning offers a promising methodology for developing skills for simulated characters, but typically requires working with sparse hand-crafted features. Building on recent progress in deep reinforcement learning (DeepRL), we introduce a mixture of actor-critic experts (MACE) approach that learns terrain-adaptive dynamic locomotion skills using high-dimensional state and terrain descriptions as input, and parameterized leaps or steps as output actions. MACE learns more quickly than a single actor-critic approach and results in actor-critic experts that exhibit specialization. Additional elements of our solution that contribute towards efficient learning include Boltzmann exploration and the use of initial actor biases to encourage specialization. Results are demonstrated for multiple planar characters and terrain classes.
Supplemental Material
Available for Download
Supplemental files.
- Assael, J.-A. M., Wahlström, N., Schön, T. B., and Deisenroth, M. P. 2015. Data-efficient learning of feedback policies from image pixels using deep dynamical models. arXiv preprint arXiv:1510.02173.Google Scholar
- Bullet, 2015. Bullet physics library, Dec. http://bulletphysics.org.Google Scholar
- Calinon, S., Kormushev, P., and Caldwell, D. G. 2013. Compliant skills acquisition and multi-optima policy search with em-based reinforcement learning. Robotics and Autonomous Systems 61, 4, 369--379. Google ScholarDigital Library
- Coros, S., Beaudoin, P., Yin, K. K., and van de Panne, M. 2008. Synthesis of constrained walking skills. ACM Trans. Graph. 27, 5, Article 113. Google ScholarDigital Library
- Coros, S., Beaudoin, P., and van de Panne, M. 2009. Robust task-based control policies for physics-based characters. ACM Transctions on Graphics 28, 5, Article 170. Google ScholarDigital Library
- Coros, S., Beaudoin, P., and van de Panne, M. 2010. Generalized biped walking control. ACM Transctions on Graphics 29, 4, Article 130. Google ScholarDigital Library
- Coros, S., Karpathy, A., Jones, B., Reveret, L., and van de Panne, M. 2011. Locomotion skills for simulated quadrupeds. ACM Transactions on Graphics 30, 4, Article 59. Google ScholarDigital Library
- da Silva, M., Abe, Y., and Popović, J. 2008. Interactive simulation of stylized human locomotion. ACM Trans. Graph. 27, 3, Article 82. Google ScholarDigital Library
- da Silva, M., Durand, F., and Popović, J. 2009. Linear bellman combination for control of character animation. ACM Trans. Graph. 28, 3, Article 82. Google ScholarDigital Library
- Doya, K., Samejima, K., Katagiri, K.-i., and Kawato, M. 2002. Multiple model-based reinforcement learning. Neural computation 14, 6, 1347--1369. Google ScholarDigital Library
- Faloutsos, P., van de Panne, M., and Terzopoulos, D. 2001. Composable controllers for physics-based character animation. In Proceedings of SIGGRAPH 2001, 251--260. Google ScholarDigital Library
- Featherstone, R. 2014. Rigid body dynamics algorithms. Springer. Google ScholarDigital Library
- Geijtenbeek, T., and Pronost, N. 2012. Interactive character animation using simulated physics: A state-of-the-art review. In Computer Graphics Forum, vol. 31, Wiley Online Library, 2492--2515. Google ScholarDigital Library
- Grzeszczuk, R., Terzopoulos, D., and Hinton, G. 1998. Neuroanimator: Fast neural network emulation and control of physics-based models. In Proc. ACM SIGGRAPH, ACM, 9--20. Google ScholarDigital Library
- Hansen, N. 2006. The cma evolution strategy: A comparing review. In Towards a New Evolutionary Computation, 75--102.Google Scholar
- Haruno, M., Wolpert, D. H., and Kawato, M. 2001. Mosaic model for sensorimotor learning and control. Neural computation 13, 10, 2201--2220. Google ScholarDigital Library
- Hausknecht, M., and Stone, P. 2015. Deep reinforcement learning in parameterized action space. arXiv preprint arXiv:1511.04143.Google Scholar
- Heess, N., Wayne, G., Silver, D., Lillicrap, T., Erez, T., and Tassa, Y. 2015. Learning continuous control policies by stochastic value gradients. In Advances in Neural Information Processing Systems, 2926--2934. Google ScholarDigital Library
- Hester, T., and Stone, P. 2013. Texplore: real-time sample-efficient reinforcement learning for robots. Machine Learning 90, 3, 385--429. Google ScholarDigital Library
- Hodgins, J. K., Wooten, W. L., Brogan, D. C., and O'Brien, J. F. 1995. Animating human athletics. In Proceedings of SIGGRAPH 1995, 71--78. Google ScholarDigital Library
- Jacobs, R. A., Jordan, M. I., Nowlan, S. J., and Hinton, G. E. 1991. Adaptive mixtures of local experts. Neural computation 3, 1, 79--87.Google Scholar
- Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., and Darrell, T. 2014. Caffe: Convolutional architecture for fast feature embedding. In Proceedings of the ACM International Conference on Multimedia, ACM, New York, NY, USA, MM '14, 675--678. Google ScholarDigital Library
- Laszlo, J., van de Panne, M., and Fiume, E. 1996. Limit cycle control and its application to the animation of balancing and walking. In Proc. ACM SIGGRAPH, 155--162. Google ScholarDigital Library
- Lee, J., and Lee, K. H. 2006. Precomputing avatar behavior from human motion data. Graphical Models 68, 2, 158--174. Google ScholarDigital Library
- Lee, Y., Lee, S. J., and Popović, Z. 2009. Compact character controllers. ACM Transctions on Graphics 28, 5, Article 169. Google ScholarDigital Library
- Lee, Y., Wampler, K., Bernstein, G., Popović, J., and Popović, Z. 2010. Motion fields for interactive character locomotion. ACM Transctions on Graphics 29, 6, Article 138. Google ScholarDigital Library
- Lee, Y., Kim, S., and Lee, J. 2010. Data-driven biped control. ACM Transctions on Graphics 29, 4, Article 129. Google ScholarDigital Library
- Levine, S., and Abbeel, P. 2014. Learning neural network policies with guided policy search under unknown dynamics. In Advances in Neural Information Processing Systems 27. 1071--1079. Google ScholarDigital Library
- Levine, S., and Koltun, V. 2014. Learning complex neural network policies with trajectory optimization. In Proceedings of the 31st International Conference on Machine Learning (ICML-14), 829--837.Google Scholar
- Levine, S., Wang, J. M., Haraux, A., Popović, Z., and Koltun, V. 2012. Continuous character control with low-dimensional embeddings. ACM Transactions on Graphics (TOG) 31, 4, 28. Google ScholarDigital Library
- Levine, S., Finn, C., Darrell, T., and Abbeel, P. 2015. End-to-end training of deep visuomotor policies. arXiv preprint arXiv:1504.00702. Google ScholarDigital Library
- Lillicrap, T. P., Hunt, J. J., Pritzel, A., Heess, N., Erez, T., Tassa, Y., Silver, D., and Wierstra, D. 2015. Continuous control with deep reinforcement learning. arXiv preprint arXiv:1509.02971.Google Scholar
- Liu, L., Yin, K., va n d e Panne, M., and Guo, B. 2012. Terrain runner: control, parameterization, composition, and planning for highly dynamic motions. ACM Trans. Graph. 31, 6, 154. Google ScholarDigital Library
- Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al. 2015. Human-level control through deep reinforcement learning. Nature 518, 7540, 529--533.Google Scholar
- Mordatch, I., and Todorov, E. 2014. Combining the benefits of function approximation and trajectory optimization. In Robotics: Science and Systems (RSS).Google Scholar
- Mordatch, I., de Lasa, M., and Hertzmann, A. 2010. Robust physics-based locomotion using low-dimensional planning. ACM Trans. Graph. 29, 4, Article 71. Google ScholarDigital Library
- Mordatch, I., Lowrey, K., Andrew, G., Popovic, Z., and Todorov, E. V. 2015. Interactive control of diverse complex characters with neural networks. In Advances in Neural Information Processing Systems, 3114--3122. Google ScholarDigital Library
- Muico, U., Lee, Y., Popović, J., and Popović, Z. 2009. Contact-aware nonlinear control of dynamic characters. ACM Trans. Graph. 28, 3, Article 81. Google ScholarDigital Library
- Muico, U., Popović, J., and Popović, Z. 2011. Composite control of physically simulated characters. ACM Trans. Graph. 30, 3, Article 16. Google ScholarDigital Library
- Nair, A., Srinivasan, P., Blackwell, S., Alcicek, C., Fearon, R., De Maria, A., Panneershelvam, V., Suley-man, M., Beattie, C., Petersen, S., et al. 2015. Massively parallel methods for deep reinforcement learning. arXiv preprint arXiv:1507.04296.Google Scholar
- Parisotto, E., Ba, J. L., and Salakhutdinov, R. 2015. Actor-mimic: Deep multitask and transfer reinforcement learning. arXiv preprint arXiv:1511.06342.Google Scholar
- Pastor, P., Kalakrishnan, M., Righetti, L., and Schaal, S. 2012. Towards associative skill memories. In Humanoid Robots (Humanoids), 2012 12th IEEE-RAS International Conference on, IEEE, 309--315.Google Scholar
- Peng, X. B., Berseth, G., and van de Panne, M. 2015. Dynamic terrain traversal skills using reinforcement learning. ACM Transactions on Graphics 34, 4. Google ScholarDigital Library
- Rusu, A. A., Colmenarejo, S. G., Gulcehre, C., Desjardins, G., Kirkpatrick, J., Pascanu, R., Mnih, V., Kavukcuoglu, K., and Hadsell, R. 2015. Policy distillation. arXiv preprint arXiv:1511.06295.Google Scholar
- Schaul, T., Quan, J., Antonoglou, I., and Silver, D. 2015. Prioritized experience replay. arXiv preprint arXiv:1511.05952.Google Scholar
- Schulman, J., Levine, S., Moritz, P., Jordan, M. I., and Abbeel, P. 2015. Trust region policy optimization. CoRR abs/1502.05477.Google Scholar
- Silver, D., Lever, G., Heess, N., Degris, T., Wierstra, D., and Riedmiller, M. 2014. Deterministic policy gradient algorithms. In ICML.Google Scholar
- Sok, K. W., Kim, M., and Lee, J. 2007. Simulating biped behaviors from human motion data. ACM Trans. Graph. 26, 3, Article 107. Google ScholarDigital Library
- Stadie, B. C., Levine, S., and Abbeel, P. 2015. Incentiviz-ing exploration in reinforcement learning with deep predictive models. arXiv preprint arXiv:1507.00814.Google Scholar
- Tan, J., Liu, K., and Turk, G. 2011. Stable proportional-derivative controllers. Computer Graphics and Applications, IEEE 31, 4, 34--44. Google ScholarDigital Library
- Tan, J., Gu, Y., Liu, C. K., and Turk, G. 2014. Learning bicycle stunts. ACM Transactions on Graphics (TOG) 33, 4, 50. Google ScholarDigital Library
- Treuille, A., Lee, Y., and Popović, Z. 2007. Near-optimal character animation with continuous control. ACM Transactions on Graphics (TOG) 26, 3, Article 7. Google ScholarDigital Library
- Uchibe, E., and Doya, K. 2004. Competitive-cooperative-concurrent reinforcement learning with importance sampling. In Proc. of International Conference on Simulation of Adaptive Behavior: From Animals and Animats, 287--296.Google Scholar
- van der Maaten, L., and Hinton, G. E. 2008. Visualizing high-dimensional data using t-sne. Journal of Machine Learning Research 9, 2579--2605.Google Scholar
- Van Hasselt, H., and Wiering, M. A. 2007. Reinforcement learning in continuous action spaces. In Approximate Dynamic Programming and Reinforcement Learning, 2007. ADPRL 2007. IEEE International Symposium on, IEEE, 272--279.Google Scholar
- Van Hasselt, H., Guez, A., and Silver, D. 2015. Deep reinforcement learning with double q-learning. arXiv preprint arXiv:1509.06461.Google Scholar
- Van Hasselt, H. 2012. Reinforcement learning in continuous state and action spaces. In Reinforcement Learning. Springer, 207--251.Google Scholar
- Wang, J. M., Fleet, D. J., and Hertzmann, A. 2009. Optimizing walking controllers. ACM Transctions on Graphics 28, 5, Article 168. Google ScholarDigital Library
- Wiering, M., and Van Hasselt, H. 2008. Ensemble algorithms in reinforcement learning. Systems, Man, and Cybernetics, Part B: Cybernetics, IEEE Transactions on 38, 4, 930--936. Google ScholarDigital Library
- Ye, Y., and Liu, C. K. 2010. Optimal feedback control for character animation using an abstract model. ACM Trans. Graph. 29, 4, Article 74. Google ScholarDigital Library
- Yin, K., Loken, K., and van de Panne, M. 2007. Simbicon: Simple biped locomotion control. ACM Transctions on Graphics 26, 3, Article 105. Google ScholarDigital Library
- Yin, K., Coros, S., Beaudoin, P., and van de Panne, M. 2008. Continuation methods for adapting simulated skills. ACM Transctions on Graphics 27, 3, Article 81. Google ScholarDigital Library
Index Terms
- Terrain-adaptive locomotion skills using deep reinforcement learning
Recommendations
Conversational Recommender System Using Deep Reinforcement Learning
RecSys '22: Proceedings of the 16th ACM Conference on Recommender SystemsDeep Reinforcement Learning (DRL) uses the best of both Reinforcement Learning and Deep Learning for solving problems which cannot be addressed by them individually. Deep Reinforcement Learning has been used widely for games, robotics etc. Limited work ...
Reward Shaping in Episodic Reinforcement Learning
AAMAS '17: Proceedings of the 16th Conference on Autonomous Agents and MultiAgent SystemsRecent advancements in reinforcement learning confirm that reinforcement learning techniques can solve large scale problems leading to high quality autonomous decision making. It is a matter of time until we will see large scale applications of ...
Deep reinforcement learning collision avoidance using policy gradient optimisation and Q-learning
Usage of trust region policy optimisation (TRPO) and proximal policy optimisation (PPO) 'children of policy gradient optimisation method' and deep Q-learning network (DQN) in Lidar-based differential robots are proposed using Turtlebot and OpenAI's ...
Comments