Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
125 tokens/sec
GPT-4o
53 tokens/sec
Gemini 2.5 Pro Pro
42 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Spatio-Temporal Motion Retargeting for Quadruped Robots (2404.11557v2)

Published 17 Apr 2024 in cs.RO

Abstract: This work introduces a motion retargeting approach for legged robots, which aims to create motion controllers that imitate the fine behavior of animals. Our approach, namely spatio-temporal motion retargeting (STMR), guides imitation learning procedures by transferring motion from source to target, effectively bridging the morphological disparities by ensuring the feasibility of imitation on the target system. Our STMR method comprises two components: spatial motion retargeting (SMR) and temporal motion retargeting (TMR). On the one hand, SMR tackles motion retargeting at the kinematic level by generating kinematically feasible whole-body motions from keypoint trajectories. On the other hand, TMR aims to retarget motion at the dynamic level by optimizing motion in the temporal domain. We showcase the effectiveness of our method in facilitating Imitation Learning (IL) for complex animal movements through a series of simulation and hardware experiments. In these experiments, our STMR method successfully tailored complex animal motions from various media, including video captured by a hand-held camera, to fit the morphology and physical properties of the target robots. This enabled RL policy training for precise motion tracking, while baseline methods struggled with highly dynamic motion involving flying phases. Moreover, we validated that the control policy can successfully imitate six different motions in two quadruped robots with different dimensions and physical properties in real-world settings.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (48)
  1. Skeleton-aware networks for deep motion retargeting. ACM Transactions on Graphics 39. URL: https://dl.acm.org/doi/10.1145/3386569.3392462, doi:10.1145/3386569.3392462.
  2. Robust Physics‐based Motion Retargeting with Realistic Body Shapes. Computer Graphics Forum 37, 81–92. URL: https://onlinelibrary.wiley.com/doi/10.1111/cgf.13514, doi:10.1111/cgf.13514.
  3. State estimation for legged robots: Consistent fusion of leg kinematics and imu .
  4. Review of the damped least-squares inverse kinematics with experiments on an industrial robot manipulator. IEEE Transactions on Control Systems Technology 2, 123–134. URL: http://ieeexplore.ieee.org/document/294335/, doi:10.1109/87.294335.
  5. Towards a natural motion generator: a pipeline to control a humanoid based on motion data, in: 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 4373–4380. doi:10.1109/IROS40897.2019.8967941.
  6. Nonparametric motion retargeting for humanoid robots on shared latent space, in: 16th Robotics: Science and Systems, RSS 2020, MIT Press Journals.
  7. Self-Supervised Motion Retargeting with Safety Guarantee, in: 2021 IEEE International Conference on Robotics and Automation (ICRA), IEEE, Xi’an, China. pp. 8097–8103. URL: https://ieeexplore.ieee.org/document/9560860/, doi:10.1109/ICRA48506.2021.9560860.
  8. Learning modular neural network policies for multi-task and multi-robot transfer, in: 2017 IEEE international conference on robotics and automation (ICRA), IEEE. pp. 2169–2176.
  9. Adversarial Motion Priors Make Good Substitutes for Complex Reward Functions, in: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, Kyoto, Japan. pp. 25–32. URL: https://ieeexplore.ieee.org/document/9981973/, doi:10.1109/IROS47612.2022.9981973.
  10. Genloco: Generalized locomotion controllers for quadrupedal robots, in: Conference on Robot Learning, PMLR. pp. 1893–1903.
  11. OPT-Mimic: Imitation of Optimized Trajectories for Dynamic Quadruped Behaviors, in: 2023 IEEE International Conference on Robotics and Automation (ICRA), IEEE, London, United Kingdom. pp. 5092–5098. URL: https://ieeexplore.ieee.org/document/10160562/, doi:10.1109/ICRA48891.2023.10160562.
  12. DOC: Differentiable Optimal Control for Retargeting Motions onto Legged Robots. ACM Transactions on Graphics 42, 1–14. URL: https://dl.acm.org/doi/10.1145/3592454, doi:10.1145/3592454.
  13. Bagail: Multi-modal imitation learning from imbalanced demonstrations. Neural Networks , 106251.
  14. Residual reinforcement learning for robot control, in: 2019 International Conference on Robotics and Automation (ICRA), pp. 6023–6029. doi:10.1109/ICRA.2019.8794127.
  15. RL + Model-Based Control: Using On-Demand Optimal Control to Learn Versatile Legged Locomotion. IEEE Robotics and Automation Letters 8, 6619–6626. URL: https://ieeexplore.ieee.org/document/10225268/, doi:10.1109/LRA.2023.3307008.
  16. Animal Motions on Legged Robots Using Nonlinear Model Predictive Control, in: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, Kyoto, Japan. pp. 11955–11962. URL: https://ieeexplore.ieee.org/document/9981945/, doi:10.1109/IROS47612.2022.9981945.
  17. Animal Gaits on Quadrupedal Robots Using Motion Matching and Model-Based Control, in: 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, Prague, Czech Republic. pp. 8500–8507. URL: https://ieeexplore.ieee.org/document/9635838/, doi:10.1109/IROS51168.2021.9635838.
  18. HumanConQuad: Human Motion Control of Quadrupedal Robots using Deep Reinforcement Learning, in: SIGGRAPH Asia 2022 Emerging Technologies, ACM, Daegu Republic of Korea. pp. 1–2. URL: https://dl.acm.org/doi/10.1145/3550471.3564762, doi:10.1145/3550471.3564762.
  19. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 .
  20. On-line motion retargetting, in: Proceedings. Seventh Pacific Conference on Computer Graphics and Applications (Cat. No.PR00293), IEEE Comput. Soc, Seoul, South Korea. pp. 32–42. URL: http://ieeexplore.ieee.org/document/803346/, doi:10.1109/PCCGA.1999.803346.
  21. Guided policy search, in: International conference on machine learning, PMLR. pp. 1–9.
  22. Learning agile skills via adversarial imitation of rough partial demonstrations, in: Conference on Robot Learning, PMLR. pp. 342–352.
  23. Ace: Adversarial correspondence embedding for cross morphology motion retargeting from human to nonhuman characters, in: SIGGRAPH Asia 2023 Conference Papers, Association for Computing Machinery, New York, NY, USA. URL: https://doi.org/10.1145/3610548.3618255, doi:10.1145/3610548.3618255.
  24. Isaac gym: High performance gpu-based physics simulation for robot learning. arXiv preprint arXiv:2108.10470 .
  25. Differential Dynamic Programming–A Unified Approach to the Optimization of Dynamic Systems* *This work was done during the author’s visit to the Division of Engineering and Applied Physics, Harvard University, and was supported by the U.S. Army Research Office, the U.S. Air Force Office of Scientific Rearch and the U.S. Office of Naval Research under the Joint Services Electronics Program by Contracts N00014-67-A-0298-0006, 0005. and 0008., in: Control and Dynamic Systems. Elsevier. volume 10, pp. 179–254. URL: https://linkinghub.elsevier.com/retrieve/pii/B9780120127108500108, doi:10.1016/B978-0-12-012710-8.50010-8.
  26. Using knowledge representation and task planning for robot-agnostic skills on the example of contact-rich wiping tasks, in: 2023 IEEE 19th International Conference on Automation Science and Engineering (CASE), IEEE. pp. 1–7.
  27. Reinforcement learning for a biped robot based on a cpg-actor-critic method. Neural networks 20, 723–735.
  28. Bayesian disturbance injection: Robust imitation learning of flexible policies for robot manipulation. Neural Networks 158, 42–58.
  29. DeepMimic: example-guided deep reinforcement learning of physics-based character skills. ACM Transactions on Graphics 37, 1–14. URL: https://dl.acm.org/doi/10.1145/3197517.3201311, doi:10.1145/3197517.3201311.
  30. Learning agile robotic locomotion skills by imitating animals, in: Robotics: Science and Systems. doi:10.15607/RSS.2020.XVI.064.
  31. AMP: adversarial motion priors for stylized physics-based character control. ACM Transactions on Graphics 40, 1–20. URL: https://dl.acm.org/doi/10.1145/3450626.3459670, doi:10.1145/3450626.3459670.
  32. Multicontact Motion Retargeting Using Whole-Body Optimization of Full Kinematics and Sequential Force Equilibrium. IEEE/ASME Transactions on Mechatronics 27, 4188–4198. URL: https://ieeexplore.ieee.org/document/9728754/, doi:10.1109/TMECH.2022.3152844.
  33. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 .
  34. Dynamic time warping algorithm review. Information and Computer Science Department University of Hawaii at Manoa Honolulu, USA 855, 40.
  35. Creature features: online motion puppetry for non-human characters, in: Proceedings of the 12th ACM SIGGRAPH/Eurographics Symposium on Computer Animation, ACM, Anaheim California. pp. 213–221. URL: https://dl.acm.org/doi/10.1145/2485895.2485903, doi:10.1145/2485895.2485903.
  36. Implementation of imitation learning using natural learner central pattern generator neural networks. Neural Networks 83, 94–108.
  37. Legged robots that keep on learning: Fine-tuning locomotion policies in the real world, in: 2022 International Conference on Robotics and Automation (ICRA), IEEE. pp. 1593–1599.
  38. Practical bayesian optimization of machine learning algorithms. Advances in neural information processing systems 25.
  39. Practical bayesian optimization of machine learning algorithms, in: Pereira, F., Burges, C., Bottou, L., Weinberger, K. (Eds.), Advances in Neural Information Processing Systems, Curran Associates, Inc. URL: https://proceedings.neurips.cc/paper_files/paper/2012/file/05311655a15b75fab86956663e1819cd-Paper.pdf.
  40. A physically-based motion retargeting filter. ACM Transactions on Graphics 24, 98–117. URL: https://dl.acm.org/doi/10.1145/1037957.1037963, doi:10.1145/1037957.1037963.
  41. Synthesis and stabilization of complex behaviors through online trajectory optimization, in: 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE, Vilamoura-Algarve, Portugal. pp. 4906–4913. URL: http://ieeexplore.ieee.org/document/6386025/, doi:10.1109/IROS.2012.6386025.
  42. Neural Kinematic Networks for Unsupervised Motion Retargetting, in: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE, Salt Lake City, UT. pp. 8639–8648. URL: https://ieeexplore.ieee.org/document/8578999/, doi:10.1109/CVPR.2018.00901.
  43. Animating non-humanoid characters with human motion data, in: Proceedings of the 2010 ACM SIGGRAPH/Eurographics Symposium on Computer Animation, Eurographics Association, Goslar, DEU. p. 169–178.
  44. Banmo: Building animatable 3d neural models from many casual videos, in: 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2853–2863. doi:10.1109/CVPR52688.2022.00288.
  45. Addressing implicit bias in adversarial imitation learning with mutual information. Neural Networks 167, 847–864.
  46. Hybrid learning mechanisms under a neural control network for various walking speed generation of a quadruped robot. Neural Networks 167, 292–308.
  47. Sim-to-real transfer in deep reinforcement learning for robotics: a survey, in: 2020 IEEE symposium series on computational intelligence (SSCI), IEEE. pp. 737–744.
  48. Distributional generative adversarial imitation learning with reproducing kernel generalization. Neural Networks 165, 43–59.
Citations (3)
View on Semantic Scholar

Summary

  • The paper introduces the STMR method that decouples spatial and temporal retargeting for transferring animal movement data onto quadruped robots.
  • It utilizes SMR to enforce kinematic constraints and TMR to adjust temporal parameters, eliminating artifacts like foot sliding.
  • Experiments demonstrate improved tracking performance over baseline methods, enabling quadrupeds to replicate complex animal behaviors robustly.

Spatio-Temporal Motion Retargeting for Quadruped Robots: An Expert Overview

The paper proposes a methodology for the retargeting of motion data onto quadruped robots, aimed at enhancing the fidelity and feasibility of transferred movements from animal behaviors to robotic systems. The method, referred to as Spatio-Temporal Motion Retargeting (STMR), is designed to bridge the morphological differences between source and target systems while ensuring that the generated motions are both kinematically and dynamically feasible.

Core Components and Approach

The proposed STMR consists of two main components: Spatial Motion Retargeting (SMR) and Temporal Motion Retargeting (TMR). Each component addresses different challenges associated with motion retargeting:

  1. Spatial Motion Retargeting (SMR): This component operates at the kinematic level. It focuses on eliminating kinematic artifacts such as foot sliding and penetration by imposing constraints on foot movement. SMR generates a full-body motion trajectory that adheres to the contact schedules of the original motion, ensuring that the essence and timing of the movements are maintained despite the change in morphology.
  2. Temporal Motion Retargeting (TMR): Working at the dynamic level, TMR optimizes the temporal parameters to ensure that the motion is dynamically feasible. It focuses on characteristics such as flight phases in jumping motions, which naturally require adjusted timing depending on the size and physical properties of the robot. The authors utilize a model-based control approach to score and iteratively refine the temporal parameters, assessing the feasibility and performance of the retargeted motion.

Numerical Results and Experiments

The paper details a series of simulation and real-world experiments to evaluate the efficacy of the STMR approach. The authors conducted extensive trials across six distinct motions and three different quadruped robotic configurations. Their method successfully preserved contact schedules and eliminated sliding artifacts, while enabling robots of various sizes to mimic six complex animal motions.

In the experiments, the authors compared the STMR-generated control policies to three baseline imitation learning methods. Tracking performance improvements were observed with significant reductions in L1 error distances, showcasing the advantage of using the STMR-generated kino-dynamically feasible reference motions. Furthermore, STMR facilitated the extraction and reconstruction of whole-body motion from videos captured with hand-held cameras, proving its capability to work with data lacking global body pose information.

Practical Implications and Future Directions

The development of STMR presents several practical benefits for the field of robotics. It offers a robust framework for adapting complex animal movements to quadruped robots, which is valuable for enhancing robotic capabilities in dynamic and interactive environments, such as search and rescue operations or assistance robots. The method also supports the integration of diverse motion data sources, enriching the repertoire of robotic skills and improving deployment versatility.

Theoretically, this work reinforces the importance of accounting for both spatial and temporal aspects in motion retargeting, demonstrating that a decoupled approach can yield feasible and expressive robotic movements. Future research directions could explore the generalization of STMR to other legged robotic morphologies, such as humanoids, and enhance robustness against environmental uncertainties and modeling discrepancies. Additionally, the application of STMR in conjunction with real-time control and perception modules could further facilitate the use of robots in complex, real-world scenarios.

Overall, this paper provides a rigorous and thoughtful contribution to the motion imitation field, emphasizing the nuanced challenges and solutions within the context of transferring detailed animal motions to robotic counterparts.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown
Youtube Logo Streamline Icon: https://streamlinehq.com

YouTube