Show Reference: "Optimality principles in sensorimotor control"

Optimality principles in sensorimotor control Nature Neuroscience, Vol. 7, No. 9. (26 August 2004), pp. 907-915, doi:10.1038/nn1309 by Emanuel Todorov
@article{todorov-2004,
    abstract = {The sensorimotor system is a product of evolution, development, learning and adaptation—which work on different time scales to improve behavioral performance. Consequently, many theories of motor function are based on 'optimal performance': they quantify task goals as cost functions, and apply the sophisticated tools of optimal control theory to obtain detailed behavioral predictions. The resulting models, although not without limitations, have explained more empirical phenomena than any other class. Traditional emphasis has been on optimizing desired movement trajectories while ignoring sensory feedback. Recent work has redefined optimality in terms of feedback control laws, and focused on the mechanisms that generate behavior online. This approach has allowed researchers to fit previously unrelated concepts and observations into what may become a unified theoretical framework for interpreting motor function. At the heart of the framework is the relationship between high-level goals, and the real-time sensorimotor control strategies most suitable for accomplishing those goals.},
    author = {Todorov, Emanuel},
    day = {26},
    doi = {10.1038/nn1309},
    issn = {1097-6256},
    journal = {Nature Neuroscience},
    keywords = {biology, math, motor},
    month = aug,
    number = {9},
    pages = {907--915},
    pmid = {15332089},
    posted-at = {2013-01-09 17:21:51},
    priority = {2},
    publisher = {Nature Publishing Group},
    title = {Optimality principles in sensorimotor control},
    url = {http://dx.doi.org/10.1038/nn1309},
    volume = {7},
    year = {2004}
}

See the CiteULike entry for more info, PDF links, BibTex etc.

Open-loop control (in biological sensorimotor modeling) is a simplification because most motions are not ballistic.

Cost terms that are routinely minimized in sensorimotor control are

  • Metabolic (muscular) energy consumption
  • smoothness cost (time-derivative of acceleration)
  • variance (of execution; usually assuming motor noise is control dependent)

Cost functions including multiple cost terms must be tuned by weighting, which is often arbitrarily done by the modeler.

Kalman filters are optimal estimators when dynamics and sensory measurements are linear and noise is Gaussian.

Explicit or implicit representations of world dynamics are necessary for optimal controllers since they have to anticipate state changes before the arrival of the necessary sensor data.

An open-loop controller is just a special case of a closed-loop controller: one that does not have feedback.

The difference between open-loop and closed-loop controllers is greatest when during control there is time to acquire sensory data and act on it.

So the question for saccade control is: is there time and where does sensory data come from?

Are there representations, forward models of saccade controls in the SC?

According to control theory, optimal controllers "make no effort to correct deviations from the average behavior", which is called the 'minimal intervention' principle:

the reason is that the effort itself is expensive and introduces control-dependent noise.

'Minimal intervention' can be used to argue liberal economics.

Todorov argues that the minimal intervention principle accounts for large variances in execution dimensions which are not task relevant and states that PCA-related analysis of behavior therefore usually finds non-relevant dimensions.