High-fidelity musculoskeletal modeling reveals that motor planning variability contributes to the speed-accuracy tradeoff

Mazen Al Borno, Saurabh Vyas, Krishna V. Shenoy, Scott L. DelpeLife 2020 · DOI

TL;DR

This paper challenges the long-held belief that the speed-accuracy tradeoff in human movement is solely due to signal-dependent motor noise. By developing a high-fidelity musculoskeletal model, the authors demonstrate that motor planning variability, arising from the difficulty of finding optimal movement solutions, can explain this tradeoff. This new theory has significant implications for understanding motor control and learning.

What Problem Does This Paper Address?

For decades, motor neuroscience has grappled with the "speed-accuracy tradeoff," a fundamental observation that faster movements tend to be less accurate, and vice versa. This phenomenon is often quantified by Fitts' Law, which relates movement time to the distance and size of a target. The dominant explanation for this tradeoff has been "signal-dependent noise" – essentially, that the control signals sent to muscles become noisier when they are larger, which is required for faster movements. This paper questions this assumption, proposing that the way the brain plans movements, rather than just the noise in executing them, plays a crucial role.

Methods

The core of this research is a novel computational model of the human upper extremity. This model is highly realistic, incorporating 47 muscles and five degrees of freedom, and is driven by an optimal control framework. Unlike previous models, this one synthesizes three-dimensional reaching movements de novo by minimizing a cost function that balances endpoint accuracy with muscle activation (as a proxy for metabolic cost).

The authors then use this model to explore the speed-accuracy tradeoff by simulating reaching movements to targets of varying sizes. They also analyze neural data from monkeys performing similar reaching tasks. A key aspect of their approach is the use of a stochastic optimizer (CMA-ES) to find movement solutions, which inherently introduces variability in the planning process.

Datasets

The study utilizes two main sources of data:

  1. Simulated Data: Generated by the authors' novel biomechanical model of the upper extremity. This model was trained and evaluated on a variety of simulated reaching tasks.
  2. Experimental Data:
    • Human Reaching Data: Collected from human subjects performing reaching tasks. This data was used to validate the model's kinematic predictions and compare speed-accuracy tradeoffs.
  3. Primate Neural Data: Neural activity from premotor and primary motor cortex of monkeys performing reaching movements to targets of different sizes. This data was used to investigate the relationship between neural variability and behavioral speed-accuracy.

The paper also references and compares its findings to existing datasets, such as those from Goldberg et al. (2014) for Fitts' Law analysis.

Key Insights and Ideas

  • Motor Planning, Not Just Execution Noise: The most significant insight is that the speed-accuracy tradeoff can emerge without signal-dependent motor noise. This suggests that the inherent difficulty in finding optimal movement plans is a primary driver.
  • The "Optimization Landscape": The authors propose that for challenging tasks (smaller targets), the "optimization landscape" for finding motor plans becomes more constrained. This reduces the number of "good" (fast and accurate) solutions, making it more likely for the system to converge on a less effective, slower plan.
  • Variability Shift: Counterintuitively, smaller targets lead to less variability in preparatory neural activity but more variability in movement duration. This supports the idea that planning variability is the key, as reduced planning options lead to more consistent but potentially less optimal outcomes.
  • Model Realism Matters: A highly realistic biomechanical model was crucial. Simplified models (e.g., torque-driven) failed to reproduce key behavioral phenomena, highlighting the importance of detailed musculoskeletal dynamics.

Key Results

  • Fitts' Law Emergence: The biomechanically realistic model successfully reproduced Fitts' Law, showing that movement duration increased with the index of difficulty (smaller targets, larger distances), even without explicit motor noise.
  • Neural Data Consistency: The analysis of monkey neural data showed that movements to smaller targets had less variability in preparatory neural states but greater variability in movement duration, aligning with the proposed theory.
  • Planning Variability Contribution: The study estimated that motor planning variability could account for over 75% of the difference in movement duration between large and small targets.
  • Model Validation: The model's simulated reaching trajectories and velocity profiles closely matched experimental data from human and primate studies.

Limitations and Future Work

  • Causal Link: The authors acknowledge that while their findings show a strong correlation, they do not establish a direct causal link between neural variability during preparation and Fitts' Law.
  • Stochasticity Proxy: The model uses stochastic optimization as a proxy for motor planning variability. The exact neurophysiological mechanisms of this variability are still unknown.
  • Reaction Time: The model does not explicitly include reaction time, leading to faster predicted movements than observed experimentally.
  • Future Directions:
    • Investigate the range of applications of this theory under different experimental conditions.
  • Explore other forms of stochastic optimization to better match experimental data.
  • Conduct future studies to explore the relative contributions of signal-dependent noise and motor planning.
  • Test the theory by comparing muscle activity patterns of "experts" vs. "non-experts" performing tasks.
  • Investigate how motor planning variability changes during motor skill learning.

Summary updated

Abstract

A long-standing challenge in motor neuroscience is to understand the relationship between movement speed and accuracy, known as the speed-accuracy tradeoff. Here, we introduce a biomechanically realistic computational model of three-dimensional upper extremity movements that reproduces well-known features of reaching movements. This model revealed that the speed-accuracy tradeoff, as described by Fitts' law, emerges even without the presence of motor noise, which is commonly believed to underlie the speed-accuracy tradeoff. Next, we analyzed motor cortical neural activity from monkeys reaching to targets of different sizes. We found that the contribution of preparatory neural activity to movement duration (MD) variability is greater for smaller targets than larger targets, and that movements to smaller targets exhibit less variability in population-level preparatory activity, but greater MD variability. These results propose a new theory underlying the speed-accuracy tradeoff: Fitts' law emerges from greater task demands constraining the optimization landscape in a fashion that reduces the number of 'good' control solutions (i.e., faster reaches). Thus, contrary to current beliefs, the speed-accuracy tradeoff could be a consequence of motor planning variability and not exclusively signal-dependent noise.
BibTeX
@inproceedings{borno2020high,
  author = {Mazen Al Borno and Saurabh Vyas and Krishna V. Shenoy and Scott L. Delp},
  title = {{High-fidelity musculoskeletal modeling reveals that motor planning variability contributes to the speed-accuracy tradeoff}},
  booktitle = {eLife},
  year = {2020},
  doi = {10.7554/elife.57021},
}

Loading related content...

Figure