Antagonistic Feedback Control of Muscle Length Changes for Efficient Involuntary Posture Stabilization

Area of Science:

• Biomechanics
• Computational Neuroscience
• Robotics

Background:

• Vertebrates exhibit involuntary posture stabilization through coordinated muscle activation.
• The precise mechanisms by which numerous muscles contribute to stabilizing fewer joints remain incompletely understood.

Purpose of the Study:

• To investigate the computational mechanisms underlying involuntary posture stabilization.
• To identify the most effective control strategy for maintaining a neutral body posture (NBP) under gravitational forces.

Main Methods:

• Development of a detailed computational human body model with 949 muscle action lines and 22 joints.
• Application of actor-critic reinforcement learning (ACRL) with two feedback control models: muscle length change (FCM-ML) and joint angle differences.
• Comparison of six control methods, including ACRL with a normalized Gaussian network (ACRL-NGN) and deep deterministic policy gradient.

Main Results:

• The ACRL-NGN combined with FCM-ML demonstrated the most efficient involuntary NBP stabilization.
• This optimal method utilized antagonistic feedback control based on muscle length change, eschewing synergy patterns or muscle categorization (flexors, extensors, etc.).

Conclusions:

• Vertebrate muscle control for posture likely operates without predefined muscle categories for specific joint actions.
• Muscles are involuntarily controlled to achieve the NBP, the most comfortable posture under gravity.
• The ACRL-NGN with FCM-ML model is a promising approach for humanoid muscle control and comfortable seat design.