Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Mosaic model for sensorimotor learning and control.

M Haruno1, D M Wolpert, M Kawato

  • 1ATR Human Information Processing Research Laboratories, Seika-cho, Soraku-gun, Kyoto 619-02, Japan.

Neural Computation
|September 26, 2001
PubMed
Summary

The MOSAIC model enhances motor control by learning multiple inverse models for diverse environments. Its new Expectation-Maximization algorithm improves learning robustness and generalization for adaptable robotic manipulation.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Paving the way for precision treatment of psychiatric symptoms with functional connectivity neurofeedback.

Translational psychiatry·2026
Same author

Exploration biases how forelimb reaches to a spatial target are learned.

bioRxiv : the preprint server for biology·2023
Same author

Faunal activity rhythms influencing early community succession of an implanted whale carcass offshore Sagami Bay, Japan.

Scientific reports·2018
Same author

Two is better than one: physical interactions improve motor performance in humans.

Scientific reports·2014
Same author

Downregulation of Ral GTPase-activating protein promotes tumor invasion and metastasis of bladder cancer.

Oncogene·2012
Same author

Internal models in the cerebellum.

Trends in cognitive sciences·2011

Area of Science:

  • Robotics
  • Computational Neuroscience
  • Machine Learning

Background:

  • Humans exhibit sophisticated motor control adaptable to uncertain environments.
  • The Modular Selection and Identification for Control (MOSAIC) model was previously proposed for motor learning.
  • MOSAIC integrates forward and inverse models for simultaneous learning and selection of controllers.

Purpose of the Study:

  • To extend and evaluate the MOSAIC architecture for motor learning and control.
  • To implement and compare gradient-descent and Expectation-Maximization (EM) algorithms for learning within MOSAIC.
  • To assess MOSAIC's ability to handle multiple object manipulation, generalization, and adaptive controller selection.

Main Methods:

  • Implemented the MOSAIC architecture with both gradient-descent and Expectation-Maximization (EM) algorithms.

Related Experiment Videos

  • Utilized simulations of an object manipulation task to evaluate the architecture's performance.
  • Tested the model's ability to generalize to novel object dynamics and adapt controller selection based on shape-dynamic pairings.

    • Main Results:

    • The EM algorithm demonstrated robustness to initial conditions and learning parameters, outperforming gradient descent.
    • Simulations confirmed MOSAIC's capacity to learn manipulation of multiple objects and switch controllers appropriately.
    • The model exhibited generalization to novel object dynamics and demonstrated on-line correction for novel shape-dynamic pairings.

    Conclusions:

    • The enhanced MOSAIC architecture, particularly with the EM algorithm, offers a robust and adaptable framework for motor learning and control.
    • The model successfully learns to generalize and select appropriate controllers, demonstrating potential for complex robotic applications.
    • MOSAIC's ability to adapt and correct in real-time highlights its promise for dynamic and uncertain environments.