Fundamental Challenges in Deep Learning for Stiff Contact Dynamics

被引:17
|
作者
Parmar, Mihir [1 ]
Halm, Mathew [1 ]
Posa, Michael [1 ]
机构
[1] Univ Penn, Grasp Lab, Philadelphia, PA 19104 USA
基金
美国国家科学基金会;
关键词
D O I
10.1109/IROS51168.2021.9636383
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Frictional contact has been extensively studied as the core underlying behavior of legged locomotion and manipulation, and its nearly-discontinuous nature makes planning and control difficult even when an accurate model of the robot is available. Here, we present empirical evidence that learning an accurate model in the first place can be confounded by contact, as modern deep learning approaches are not designed to capture this non-smoothness. We isolate the effects of contact's non-smoothness by varying the mechanical stiffness of a compliant contact simulator. Even for a simple system, we find that stiffness alone dramatically degrades training processes, generalization, and data-efficiency. Our results raise serious questions about simulated testing environments which do not accurately reflect the stiffness of rigid robotic hardware. Significant additional investigation will be necessary to fully understand and mitigate these effects, and we suggest several avenues for future study.
引用
收藏
页码:5181 / 5188
页数:8
相关论文
共 50 条
  • [31] Deep learning of vehicle dynamics
    Szecsi, M.
    Gyorok, B.
    Weinhardt-Kovacs, A.
    Beintema, G. I.
    Schoukens, M.
    Peni, T.
    Toth, R.
    IFAC PAPERSONLINE, 2024, 58 (15): : 283 - 288
  • [32] Toward Deep Digital Contact Tracing: Opportunities and Challenges
    Cherini, Renato
    Detke, Ramiro
    Fraire, Juan
    Madoery, Pablo G.
    Finochietto, Jorge M.
    IEEE PERVASIVE COMPUTING, 2023, 22 (04) : 15 - 25
  • [33] Simultaneous Learning of Contact and Continuous Dynamics
    Bianchini, Bibit
    Halm, Mathew
    Posa, Michael
    CONFERENCE ON ROBOT LEARNING, VOL 229, 2023, 229
  • [34] Deep Learning for Wireless Dynamics
    Lee, Heunchul
    Jeong, Jaeseong
    Wang, Zhao
    2022 IEEE INTERNATIONAL CONFERENCE ON COMMUNICATIONS WORKSHOPS (ICC WORKSHOPS), 2022, : 1141 - 1146
  • [35] Ethical challenges of machine learning and deep learning algorithms
    Prabhu, Sanjay P.
    LANCET ONCOLOGY, 2019, 20 (05): : 621 - 622
  • [36] Latent space dynamics learning for stiff collisional-radiative models
    Xie, Xuping
    Tang, Qi
    Tang, Xianzhu
    MACHINE LEARNING-SCIENCE AND TECHNOLOGY, 2024, 5 (04):
  • [37] Shared Data and Algorithms for Deep Learning in Fundamental Physics
    Benato L.
    Buhmann E.
    Erdmann M.
    Fackeldey P.
    Glombitza J.
    Hartmann N.
    Kasieczka G.
    Korcari W.
    Kuhr T.
    Steinheimer J.
    Stöcker H.
    Plehn T.
    Zhou K.
    Computing and Software for Big Science, 2022, 6 (1)
  • [38] A linearly implicit trapezoidal method for integrating stiff multibody dynamics with contact, joints, and friction
    Potra, Florian A.
    Anitescu, Mihai
    Gavreal, Bogdan
    Trinkle, Jeff
    INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, 2006, 66 (07) : 1079 - 1124
  • [39] Learning Dynamics and Generalization in Deep Reinforcement Learning
    Lyle, Clare
    Rowland, Mark
    Dabney, Will
    Kwiatkowksa, Marta
    Gal, Yarin
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 162, 2022,
  • [40] Dynamics of stiff chain
    Zhang, LX
    Xia, A
    Zhao, DL
    EUROPEAN POLYMER JOURNAL, 2001, 37 (06) : 1277 - 1281