Brian ASB Abstract - Githubissues

Title: Neuromechanical implications of postural changes to motor learning and performance Link: cohn_jalaleddini_valerocuevas_asb_2017.pdf cohn_jalaleddini_valerocuevas_asb_2017.pdf
Abstract
METHODS
We firmly connected the index fingertip of a Utah/M.I.T hand [2] to a 6-DOF load cell to produce static forces (Fig 1). The load cell was affixed to the endpoint of an AdeptSix300 robot that was moved to change finger posture. Seven index finger tendons were actuated by DC brushless motors [3], routed through pulleys.
The robot moved the limb endpoint to 100 randomly selected endpoint postures on an arc (Fig 2). At each posture, the motors applied 100 force combinations across tendons uniformly at random (spanning 3 to 12N range). The duration of each trial was 0.8s— sufficient for forces to settle. We sampled fingertip and tendon forces at 1kHz.
were recorded as motors produced known tendon forces, at different postures. The resulting endpoint force vectors (red) were described in spherical coordinates (rho, theta, phi) in the common frame of reference of the fingertip and sensor.
We calculated the force steady-state of each trial by averaging the last 0.2s. For each posture, we identified the linear static 3x7 model ( Ai  ) that transforms tendon tensions to endpoint forces using linear regression:
Note this mapping does not consider torques at the endpoint of the finger [1] and serves as a worst-case scenario for model performance.
RESULTS AND DISCUSSION
For all individual postures, a linear model  Ai, accurately predicted endpoint force as a function of tendon forces, (i.e., a high percentage of variance-accounted-for, %VAF, Fig 2). In addition, the negligible residual error did not have a structure
Forces at the tip of the mechanical finger across posture (Fig 2a). As could be expected given the nonlinear changes in the finger’s Jacobian [1], posture had a profound effect on the  Ai  matrices that map tension to endpoint force. Interestingly, the effect of posture in fingertip force strength (rho, in N) and direction differed widely across muscles. While m 2 had a consistent rho across all postures, m 3 had higher variability(Fig3, left).As for direction,m 2’ s direction in the zy plane (phi) was consistent across postures—m 3  was more variable (Fig 3, right). We conclude that linear models (i.e.,  Ai matrices) do not perform uniformly well across postures. Yet
effective neural control of tendon driven limbs should work well across the workspace [4]. Interestingly, our results suggest small changes in posture can lead to large changes in the mechanical actions of muscles—therefore  Ai  matrices likely do not generalize well across regions of the workspace. We speculate that the full mechanical output of the limb should be considered (i.e., endpoint torque output),and that exploration of the full workspace is preferable as interpolation will likely not work. Thus, the mechanical structure of the tendinous apparatus can influence motor learning. Moreover, disruption of learning or recall of these mappings can easily lead to motor pathologies.```
Author
`Cohn BA, Jalaleddini K, Valero-Cuevas FJ`
BibTex
```@inproceedings
"@article{cohn2017-asb-neuromechanical-posture,
  title={Neuromechanical implications of postural changes to motor learning and performance},
  author={Cohn BA, Jalaleddini K, Valero-Cuevas FJ},
booktitle={Proceedings of the 41st Annual Meeting of the American Society of Biomechanics, Denver, CO},
  volume={},
  year={2017}
}
Year: 2017 Journal: American Society of Biomechanics SupplementalLink:
usc-bbdl / usc-bbdl.github.io

Brian ASB Abstract #37
