This paper presents a novel framework for learning orbitally stable nonlinear dynamical systems from demonstrations for rhythmic tasks in robotics. The core innovation is a reproducing kernel Hilbert space (RKHS) parametrization method for rhythmic dynamics modeling, ensuring the existence of stable closed-loop orbits within the generated trajectories. By leveraging transverse contraction theory, we provide theoretical guarantees for the orbital stability of the learned dynamics. To address computational inefficiencies associated with linear matrix inequalities (LMI) constraints, we relax the semi-infinite constraints and simplify the parametrization, transforming the problem into iterative solutions of convex quadratic optimization problems, which can be efficiently solved. We validate the proposed algorithm through simulations and a series of real-world rhythmic tasks. The simulation results indicate that our method significantly outperforms existing approaches in accurately replicating demonstrated behaviors. Additionally, real-world experiments consistently show high performance in completing rhythmic tasks, demonstrating the method’s potential to address challenges in reproducing periodic movements and advancing rhythmic motion replication.
AhbeEWoodTASmithRS (2018) Transverse contraction-based stability analysis for periodic trajectories of controlled power kites with model uncertainty. IEEE Conference on Decision and Control. IEEE.
2.
AlvarezMARosascoLLawrenceND (2012) Kernels for vector-valued functions: a review. Foundations and Trends® in Machine Learning4(3): 195–266.
3.
ArgallBDChernovaSVelosoM, et al. (2009) A survey of robot learning from demonstration. Robotics and Autonomous Systems57: 469–483.
4.
BaddooPJHerrmannBMcKeonBJ, et al. (2022) Kernel learning for robust dynamic mode decomposition: linear and nonlinear disambiguation optimization. Proceedings of the Royal Society A478(2260): 20210830.
5.
BeckATetruashviliL (2013) On the convergence of block coordinate descent type methods. SIAM Journal on Optimization23(4): 2037–2060.
6.
Beik-MohammadiHHaubergSArvanitidisG, et al. (2024) Neural contractive dynamical systemsInternational Conference on Learning Representations.
7.
BorgG (1960) A Condition for the Existence of Orbitally Stable Solutions of Dynamical Systems. Stockholm: Elander, Vol. 153.
8.
ChenRTQRubanovaYBettencourtJ, et al. (2018) Neural ordinary differential equations. Proceedings of Conference on Neural Information Processing Systems. Advances in Neural Information Processing Systems.
9.
ChiCXuZFengS, et al. (2023) Diffusion policy: visuomotor policy learning via action diffusion. The International Journal of Robotics ResearchOnline ahead of print. DOI: 10.1177/02783649241273668.
10.
FlashTHoganN (1985) The coordination of arm movements: an experimentally confirmed mathematical model. The Journal of neuroscience: The Official Journal of the Society for Neuroscience5(7): 1688–1703.
11.
FlorencePLynchCZengA, et al. (2022) Implicit behavioral cloning. In: Conference on Robot Learning. PMLR, 158–168.
12.
GamsAPetricTNemecB, et al. (2014) Learning and adaptation of periodic motion primitives based on force feedback and human coaching interaction. IEEE-RAS International Conference on Humanoid Robots. IEEE, 166–171.
13.
Har-PeledSMazumdarS (2004) On coresets for k-means and k-median clustering. Proceedings of the thirty-sixth Annual ACM Symposium on Theory of Computing. Association for Computing Machinery, 291–300.
14.
HauserJChungCC (1994) Converse Lyapunov functions for exponentially stable periodic orbits. Systems & Control Letters23(1): 27–34.
HouZMaTWangW, et al. (2025) Contextual policy search for task-level adaptation in physical human–robot interaction. IEEE/ASME Transactions on Mechatronics. IEEE.
17.
HuSMendoncaRJohnsonMJ, et al. (2021) Robotics for occupational therapy: learning upper-limb exercises from demonstrations. IEEE Robotics and Automation Letters6(4): 7781–7788.
18.
HuangYRozoLSilvérioJ, et al. (2019) Kernelized movement primitives. The International Journal of Robotics Research38(7): 833–852.
19.
IjspeertAJNakanishiJHoffmannH, et al. (2013) Dynamical movement primitives: learning attractor models for motor behaviors. Neural Computation25: 328–373.
20.
JaffeSDavydovALapsekiliD, et al. (2024). Learning neural contracting dynamics: extended linearization and global guarantees. arXiv preprint arXiv:2402.08090.
21.
JiangSHCKrauthgamerRLouJ, et al. (2024) Coresets for kernel clustering. Machine Learning113: 1–16.
22.
KhadivarFLauzanaIBillardA (2021) Learning dynamical systems with bifurcations. Robotics and Autonomous Systems136: 103700.
23.
Khansari-ZadehSMBillardA (2011) Learning stable nonlinear dynamical systems with Gaussian mixture models. IEEE Transactions on Robotics27: 943–957.
24.
Khansari-ZadehSMBillardA (2014) Learning control Lyapunov function to ensure stability of dynamical system-based robot reaching motions. Robotics and Autonomous Systems62: 752–765.
25.
KulićDTakanoWNakamuraY (2008) Incremental learning, clustering and hierarchy formation of whole body motion patterns using adaptive hidden markov chains. The International Journal of Robotics Research27(7): 761–784.
26.
LaurainVTóthRPigaD, et al. (2020) Sparse RKHS estimation via globally convex optimization and its application in LPV-IO identification. Automatica115: 108914.
27.
LinotAJGrahamMD (2022) Data-driven reduced-order modeling of spatiotemporal chaos with neural ordinary differential equations. Chaos (Woodbury, N.Y.)32(7): 073110.
28.
LópezMStillG (2007) Semi-infinite programming. European Journal of Operational Research180(2): 491–518.
29.
ManchesterIRSlotineJJE (2014) Transverse contraction criteria for existence, stability, and robustness of a limit cycle. Systems & Control Letters63: 32–38.
30.
MarinhoZDraganAByravanA, et al. (2016) Functional gradient motion planning in reproducing kernel Hilbert spaces. arXiv preprint arXiv:1601.03648.
31.
MicchelliCAPontilM (2005) On learning vector-valued functions. Neural Computation17(1): 177–204.
32.
NawazFLiTMatniN, et al. (2024) Learning complex motion plans using neural ODEs with safety and stability guarantees. IEEE International Conference on Robotics and Automation. IEEE, 17216–17222.
33.
PapageorgiouDArgiropoulosDEDoulgeriZ (2022) Dirichlet-based dynamic movement primitives for encoding periodic motions with predefined accuracy. IEEE International Conference on Robot and Human Interactive Communication. IEEE, 1146–1152.
34.
PhillipsJMTaiWM (2018) Improved coresets for kernel density estimates. Proceedings of the Twenty-Ninth Annual ACM-SIAM Symposium on Discrete Algorithms. Society for Industrial and Applied Mathematics, 2718–2727.
35.
PhillipsJMTaiWM (2020) Near-optimal coresets of kernel density estimates. Discrete & Computational Geometry63: 867–887.
36.
RanaMALiAFoxD (2020) Euclideanizing flows: diffeomorphic reduction for learning stable dynamical systems. Proceedings of 2nd Conference on Learning for Dynamics and Control. Proceedings of Machine Learning Research.
37.
RappPE (1987) Why are so many biological systems periodic?Progress in Neurobiology29(3): 261–273.
38.
RavichandarHSalehiIDaniA (2017) Learning partially contracting dynamical systems from demonstrations. Conference on Robot Learning. Proceedings of Machine Learning Research.
39.
RavichandarHPolydorosASChernovaS, et al. (2020) Recent advances in robot learning from demonstration. Annual Review of Control Robotics and Autonomous Systems3: 297–330.
40.
SaverianoMAbu-DakkaFJKrambergerA, et al. (2023) Dynamic movement primitives in robotics: a tutorial survey. The International Journal of Robotics Research42(13): 1133–1184.
41.
SchaalS (1999) Is imitation learning the route to humanoid robots?Trends in Cognitive Sciences3(6): 233–242.
42.
SchölkopfBSmolaAJ (2001) Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond. MA: MIT press.
43.
ShafiullahNMCuiZJAltanzayaA, et al. (2022) Behavior transformers: cloning k modes with one stone. Advances in Neural Information Processing Systems35: 22955–22968.
SinghSSindhwaniVSlotineJJE, et al. (2018) Learning stabilizable dynamical systems via control contraction metrics. International Workshop on the Algorithmic Foundations of Robotics. Springer International Publishing.
46.
SinghSRichardsSMSindhwaniV, et al. (2021) Learning stabilizable nonlinear dynamics with contraction-based regularization. The International Journal of Robotics Research40(10-11): 1123–1150.
47.
SunDJhaSFanC (2021) Learning certified control using contraction metric. In: Conference on Robot Learning. ■■■.
48.
TangJZManchesterIR (2014) Transverse contraction criteria for stability of nonlinear hybrid limit cycles. IEEE Conference on Decision and Control. IEEE.
49.
UrainJGinesiMTateoD, et al. (2020) Imitationflow: learning deep stable stochastic dynamic systems by normalizing flows. IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 5231–5237.
50.
ZhangYChengLLiHC, et al. (2022) Learning accurate and stable point-to-point motions: a dynamic system approach. IEEE Robotics and Automation Letters7: 1510–1517.
51.
ZhangYChengLCaoR, et al. (2023a) A neural network based framework for variable impedance skills learning from demonstrations. Robotics and Autonomous Systems160: 104312.
52.
ZhangHYChengLZhangY (2023b) Learning robust point-to-point motions adversarially: a stochastic differential equation approach. IEEE Robotics and Automation Letters8: 2357–2364.
53.
ZhengYPhillipsJM (2017) Coresets for kernel regression. Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery, 645–654.
54.
ZhiWTangHZhangT, et al. (2024) Teaching periodic stable robot motion generation via sketch. IEEE Robotics and Automation Letters. IEEE.
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
