A biologically inspired neural network controller for ballistic arm movements

Area of Science:

• Robotics and Artificial Intelligence
• Neuroscience and Biomechanics
• Computational Motor Control

Background:

• Human arm movement involves complex sensory integration, sensorimotor transformations, and motor planning.
• Computational models, particularly Artificial Neural Networks (ANNs), are valuable tools for understanding motor control mechanisms.
• This work introduces an ANN approach to model upper limb motor control during planar ballistic movements.

Purpose of the Study:

• To develop and present a neural network model for simulating human arm motor control.
• To investigate the internal inverse model, trajectory formation, and muscular synergy generation in arm movements.
• To validate the model's performance against experimental human kinematic data.

Main Methods:

• A system comprising three computational blocks: an inverse model learning scheme, a pulse generator for muscular synergies, and a two-joint, six-actuator limb model.
• A neural controller learning paradigm based on exploration without feedback signals.
• Comparison of simulated kinematics with existing human experimental data.

Main Results:

• The model successfully reproduces human-like arm movement kinematics, including bell-shaped wrist velocity profiles and near-straight trajectories.
• Generated muscular synergies facilitate movement execution, with amplitude and direction errors of 0.52 cm and 0.2 radians, respectively.
• The neural controller demonstrates adaptability to environmental changes, such as external force fields.

Conclusions:

• The proposed neural network effectively simulates internal model development and controls ballistic planar arm movements.
• The controller's ability to learn from kinematic information and arm characteristics suggests potential applications in neuroprosthetics.
• This model could pave the way for novel rehabilitation techniques by commanding neuroprosthetic devices.