Unifying Speed-Accuracy Trade-Off and Cost-Benefit Trade-Off in Human Reaching Movements

Luka Peternel, Olivier Sigaud, Jan BabičFrontiers in Human Neuroscience 2017 · DOI

TL;DR

This paper introduces a novel computational model that unifies two fundamental principles of motor control: the speed-accuracy trade-off (SAT) and the cost-benefit trade-off (CBT). By modeling the brain as maximizing expected utility, the model successfully explains how humans balance movement speed, accuracy, and effort, a feat not achieved by previous isolated models.

What Problem Does This Paper Address?

When we reach for an object, our brain faces a complex decision-making process. Two key trade-offs are at play:

  1. Cost-Benefit Trade-Off (CBT): Moving faster towards a more rewarding target comes with increased muscular effort. The brain must balance the desire for a quick reward against the energy cost of a rapid movement.
  2. Speed-Accuracy Trade-Off (SAT): Faster movements tend to be less accurate, increasing the chance of missing the target. The brain must decide how much time to spend to achieve the desired level of accuracy.

While these trade-offs have been studied extensively, they've largely been investigated in isolation. This leaves a gap in our understanding of how the brain simultaneously manages these competing demands in real-world movements. This paper aims to bridge this gap by proposing a unified model.

Methods

The core of the paper is a new computational model that frames motor control as an expected utility maximization problem. The central nervous system (CNS) is hypothesized to optimize a cost function that considers:

  • Discounted Reward: The value of the reward decreases with movement time.
  • Probability of Hit: This is influenced by movement accuracy, which in turn depends on motor noise and target size.
  • Movement Cost: Primarily muscular effort, which increases with movement speed.

The model uses a stochastic optimal control framework, incorporating motor noise and imperfect state estimation, which are crucial for explaining variability in human movements. Specifically, they implement:

  • A two-degrees-of-freedom (DoFs) planar arm model: Controlled by six simulated muscles.
  • Signal-dependent motor noise: Mimicking biological variability.
  • Delayed feedback and state estimation: To account for the brain's need to estimate the arm's position based on noisy and delayed sensory information.
  • An optimization algorithm (CMA-ES): To tune the parameters of a neural network controller that generates muscle activations to maximize the expected utility.

A key innovation is the inclusion of a hit dispersion term in the cost function, directly linking motor noise to the probability of missing the target. They also introduce a perpendicularity cost to better match experimental data on how subjects orient their movements.

Datasets

The paper uses two main sources of data:

  1. Simulated Data: Generated by their computational model, which was trained and evaluated using the methods described above.
  2. Experimental Data: Collected from ten healthy male volunteers performing reaching movements in a horizontal plane using a haptic manipulator. This setup allowed for precise recording of hand trajectories and simulated viscous resistance to amplify movement cost.

The experimental design involved varying target sizes and starting positions to probe the interplay of CBT and SAT. The data collected included movement trajectories, movement times, velocity profiles, and hit dispersion.

Key Insights and Ideas

  • Unified Optimization: The most significant insight is that a single optimization principle (maximizing expected utility, including the probability of hitting) can explain both speed-accuracy and cost-benefit trade-offs. This suggests a more integrated neural strategy than previously assumed.
  • Stochasticity is Key: Explicitly modeling motor noise and imperfect state estimation is crucial for reproducing Fitts' Law and hit dispersion patterns, highlighting the brain's adaptation to inherent biological variability.
  • Hit Dispersion as a Predictor: The model's ability to reproduce the distribution of where subjects hit the target (hit dispersion) is a novel and distinguishing feature, suggesting that the brain actively manages this variability.
  • Asymmetric Velocity Profiles: The model's ability to generate asymmetric bell-shaped velocity profiles, where movements slow down towards the end, is explained by the need for better state estimation when accuracy is paramount.

Key Results

  • Qualitative Match: The model qualitatively reproduces key experimental findings:
  • CBT: Velocity increases with target distance, indicating the brain accepts higher costs for longer movements.
  • SAT: Movement time increases as target size decreases (consistent with Fitts' Law), showing the trade-off between speed and accuracy.
  • Hit Dispersion: The model accurately predicts the distribution of hits, showing smaller dispersion for smaller targets, reflecting the brain's effort to increase accuracy when needed.
  • Fitts' Law: The model's simulated movements adhere to Fitts' Law (r² > 0.7), similar to experimental data (r² > 0.9).
  • Movement Times: While the model's movement times were generally faster than subjects' (attributed to experimental setup like haptic damping), the relationships between time, distance, and target size were qualitatively similar.

Limitations and Future Work

  • Movement Time Discrepancy: The most noted limitation is the faster movement times in the model compared to human subjects. This is attributed to factors like the haptic manipulator's external damping and potential inaccuracies in the arm model or state estimation, which were simplified for generality.
  • Model Simplifications: The model did not explicitly account for external damping or muscular friction, which could influence movement times.
  • Two-Component Strategy: The model doesn't explicitly capture the observed two-component reaching strategy (a rapid initial phase followed by a slower approach phase) that some studies suggest.
  • Future Directions:
    • Parameter Tuning: Further exploration of meta-parameter sensitivity and optimization processes.
  • Motor Learning: Shifting focus to motor learning by studying trajectory evolution during training.
  • Discounting Models: Investigating different reward discounting functions (linear, hyperbolic, sigmoidal).
  • Multi-Component Strategy: Extending the model to incorporate multi-component reaching strategies.
  • Flat Targets: Further accounting for target orientation in the model.

Summary updated

Abstract

Two basic trade-offs interact while our brain decides how to move our body. First, with the cost-benefit trade-off, the brain trades between the importance of moving faster toward a target that is more rewarding and the increased muscular cost resulting from a faster movement. Second, with the speed-accuracy trade-off, the brain trades between how accurate the movement needs to be and the time it takes to achieve such accuracy. So far, these two trade-offs have been well studied in isolation, despite their obvious interdependence. To overcome this limitation, we propose a new model that is able to simultaneously account for both trade-offs. The model assumes that the central nervous system maximizes the expected utility resulting from the potential reward and the cost over the repetition of many movements, taking into account the probability to miss the target. The resulting model is able to account for both the speed-accuracy and the cost-benefit trade-offs. To validate the proposed hypothesis, we confront the properties of the computational model to data from an experimental study where subjects have to reach for targets by performing arm movements in a horizontal plane. The results qualitatively show that the proposed model successfully accounts for both cost-benefit and speed-accuracy trade-offs.
BibTeX
@inproceedings{peternel2017unifying,
  author = {Luka Peternel and Olivier Sigaud and Jan Babič},
  title = {{Unifying Speed-Accuracy Trade-Off and Cost-Benefit Trade-Off in Human Reaching Movements}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2017},
  doi = {10.3389/fnhum.2017.00615},
}

Loading related content...

Figure