Reconstructing State-Space from Movie Using Convolutional Autoencoder for Robot Control

  • Kazuma Takahara
  • Shuhei IkemotoEmail author
  • Koh Hosoda
Conference paper
Part of the Advances in Intelligent Systems and Computing book series (AISC, volume 867)


In contrast with intensive studies for hardware development in soft robotics, approaches to construct a controller for soft robots has been relying on heuristics. One of the biggest reasons of this issue is that even reconstructing the state-space to describe the behavior is difficult. In this study, we propose a method to reconstruct state-space from movies using a convolutional autoencoder for robot control. In the proposed method, the process that reduces the number of dimensions of each frame in movies is regulated by additional losses making latent variables orthogonal each other and apt to model the forward dynamics. The proposed method was successfully validated through a simulation where a two links planar manipulator is modeled using the movie and controlled based on the forward model.


State-space reconstruction Convolutional autoencoder 


  1. [Trimmer2014]
    Trimmer, B.: A journal of soft robotics: why now? Soft Rob. 1(1), 1–4 (2014)CrossRefGoogle Scholar
  2. [Hosoda2012]
    Hosoda, K., Sekimoto, S., Nishigori, Y., Takamuku, S., Ikemoto, S.: Anthropomorphic muscular-skeletal robotic upper limb for understanding embodied intelligence. Adv. Rob. 26(7), 729–744 (2012)CrossRefGoogle Scholar
  3. [Peters2006]
    Peters, J., Schaal, S.: Policy gradient methods for robotics. In: IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 2219–2225 (2006)Google Scholar
  4. [Akrour2016]
    Akrour, R., Neumann, G., Abdulsamad, H., et al.: Model-free trajectory optimization for reinforcement learning. In: International Conference on Machine Learning, pp. 2961–2970 (2016)Google Scholar
  5. [Soloway1996]
    Soloway, D., Pamela, J.H.: Neural generalized predictive control. Intell. Control, 277–282 (1996)Google Scholar
  6. [Wahlstrom2015]
    Wahlstrom, N., Schon, T.B., Deisenroth, M.P.: Learning deep dynamical models from image pixels. IFAC-PapersOnLine 48(28), 1059–1064 (2015)CrossRefGoogle Scholar
  7. [Watter2015]
    Watter, M., Springenberg, J., Boedecker, J., et al.: Embed to control: a locally linear latent dynamics model for control from raw images. In: Advances in Neural Information Processing Systems, pp. 2746–2754 (2015)Google Scholar
  8. [Lenz2015]
    Lenz, I., Knepper, R.A., Saxena, A.: DeepMPC: learning deep latent features for model predictive control. Rob. Sci. Syst. (2015)Google Scholar
  9. [Berniker2015]
    Berniker, M., Kording, K.P.: Deep networks for motor control functions. Frontiers Comput. Neurosci. 9 (2015)Google Scholar
  10. [Wang2016]
    Wang, M., Li, H.X., Shen, W.: Deep auto-encoder in model reduction of lage-scale spatiotemporal dynamics. In: International Joint Conference on Neural Networks (IJCNN), pp. 3180–3186. IEEE (2016)Google Scholar
  11. [Kashima2016]
    Kashima, K.: Nonlinear model reduction by deep autoencoder of noise response data. In: IEEE 55th Conference on Decision and Control (CDC), pp. 5750–5755. IEEE (2016)Google Scholar
  12. [Soter2017]
    Soter, G., Conn, A., Hauser, H., Rossiter, J.: Bodily aware soft robots: integration of proprioceptive and exteroceptive sensors. arXiv:1710.05419v2 [cs.RO], 8 November 2017
  13. [Ikemoto2018]
    Ikemoto, S., Duan, Y., Takahara, K., Kumi, T., Hosoda, K.: Robot control based on analytical models extracted from a neural network. In: The 1st International Symposium on Systems Intelligence Division (2018)Google Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  1. 1.Osaka UniversityToyonaka, OsakaJapan

Personalised recommendations