Skip to main content
  • Poster presentation
  • Open access
  • Published:

The number of granular cells in a cerebellar neuronal network model engaged during robot control increases with the complexity of the motor task


The granular cells (Gr) in the cerebellum have been proposed to perform adaptive spatio-temporal coding [1] to relay rich information to the Purkinje cell, and thus they are critical for the proper operation of the cerebellum [2]. Elimination of all Gr cells induced behavioral symptoms such as ataxia and hypotonia [3]. Partially silencing Gr cells showed that the overall motor performance was minimally affected whereas demanding tasks or memory consolidation processes were compromised [4]. Here we use a physio-anatomically inspired cerebellar neuronal network (CNN) to study the role of the Gr cells during control of an unstable two-wheel balancing robot [5].


The CNN comprises 755 Gr, 5 Golgi (Go), 15 basket/stellate (Ba/St), and 1 Purkinje (Pk) cells. Inputs to the CNN carried by mossy fibers (mfs) provided the desired motion trajectories of the robot. Inhibitory feedback loop between Gr and Go, and feed-forward inhibitory loop between Ba/St and Pk were included as in the real cerebellum. A proportional and differential (PD) controller sharing the same mfs inputs was introduced to represent the non-cerebellar input to the vestibular nucleus (Vn). The Vn computed the arithmetic difference between the PD and the Pk output and produced the motor command to drive the robot. The error signal to the Pk cell is conveyed by the climber fiber (cf). Bidirectional plasticity at Gr-Pk synapses (LTP/LTD) induced by the cf was implemented as a basis for motor learning [6]. The desired motion of the wheel angle position of the robot was sinusoidal waves or a band limited random noise (BLRN) while that of the body tilt angle was zero. The complexity of the control task changed from simple to difficult as the frequency of the sinusoidal wave changed form 0.1 Hz to 0.4 Hz, and very difficult when the BLRN (cutoff frequencies 0.2 Hz and 0.5 Hz) desired motion was used. The sinusoidal desired motion was given for 100 cycles and the BLRN for 10 s. At the end of the experiment the Gr-Pk synaptic weights were sorted to identify the number of cells that contributed to the 90% of the input to the Pk cell.


It was found that less than half of the Gr cells in the CNN were required to form 90% of the Pk output for all the desired motion employed. Interestingly, as the complexity of the motor task increased, the number of Gr cells required also increased. Knocking down the output of the most contributing 100 Gr cells showed that the motor performance was momentarily affected, but the CNN was able to re-adjust the other Gr-Pk synaptic weights and recover the motor performance. These results are in line with other computational models, such as [7] that predicted that motor performance would be sustained despite using only a portion of Gr cells.


In our robot control framework, the number of Gr cells driving the Pk cell output increased with the complexity of the motor task. Sparse coding in the Gr layer permitted the engagement of Gr cells to foster the motor performance.


  1. Solinas S, et al: A realistic large-scale model of the cerebellum granular layer predicts circuit spatio-temporal filtering properties. Front Cell Neurosci. 2012, 4 (12):

  2. D'Angelo E: Rebuilding cerebellar network computations from cellular neurophysiology. Front Cell Neurosc. 2010, 4 (131):

  3. Kim JC, et al: Linking genetically defined neurons to behavior through a broadly applicable silencing allele. Neuron. 2009, 63 (3): 305-15. 10.1016/j.neuron.2009.07.010.

    Article  PubMed Central  CAS  PubMed  Google Scholar 

  4. Galliano E, et al: Silencing the majority of cerebellar granule cells uncovers their essential role in motor learning and consolidation. Cell Rep. 2013, 3 (4): 1239-51. 10.1016/j.celrep.2013.03.023.

    Article  CAS  PubMed  Google Scholar 

  5. Pinzon-Morales R-D, et al: Error signals used for cerebellar motor learning: evaluation in real world adaptive robot control using a cerebellar neuronal network model. J Neural Eng. (in review)

  6. Ito M: Cerebellar circuitry as a neuronal machine. Prog Neurobiol. 2006, 78 (3-5): 272-303. 10.1016/j.pneurobio.2006.02.006.

    Article  PubMed  Google Scholar 

  7. Schweighofer N, et al: Unsupervised learning of granule cell sparse codes enhances cerebellar adaptive control. Neuroscience. 2001, 103 (1): 35-50. 10.1016/S0306-4522(00)00548-0.

    Article  CAS  PubMed  Google Scholar 

Download references

Author information

Authors and Affiliations


Corresponding author

Correspondence to Yutaka Hirata.

Rights and permissions

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Pinzon-Morales, RD., Hirata, Y. The number of granular cells in a cerebellar neuronal network model engaged during robot control increases with the complexity of the motor task. BMC Neurosci 15 (Suppl 1), P143 (2014).

Download citation

  • Published:

  • DOI: