Modeling sensorimotor control of human upright stance

Area of Science:

• Biomedical Engineering
• Neuroscience
• Robotics

Background:

• Human postural control relies on integrating multisensory data to estimate external disturbances.
• Current engineering models often use complex 'internal observer' systems for multisensory estimation.

Purpose of the Study:

• To model human postural control using simple sensor fusion mechanisms.
• To compare simple sensor fusion with complex internal observer models for estimating disturbances.
• To validate findings in simulations and on a biped humanoid robot.

Main Methods:

• Developed a model using weighted sums of sensory signals combined with thresholds for disturbance estimation.
• Simulated human postural behavior under various conditions.
• Implemented and tested the model on a biped humanoid robot with real-world sensors.

Main Results:

• Simple sensor fusion mechanisms effectively mimic human-like postural behavior in simulations.
• The model demonstrated robustness in real-world robotic applications with noisy sensors.
• Complex internal observer models also produced human-like bipedal posture control.

Conclusions:

• Humans likely employ both simple, hardwired sensor fusion for automatic reactions and complex internal observers for voluntary movements.
• Sensor fusion with thresholding is evolutionarily optimized, while internal observers facilitate sensorimotor learning.