Predicting fall risk using multiple mechanics-based metrics for a planar biped model.

Area of Science:

• Biomechanics and Robotics
• Gait Stability Analysis
• Predictive Modeling

Background:

• Falls are a significant concern for both humans and robotic systems, necessitating the development of robust fall prediction models.
• Numerous mechanics-based fall risk metrics exist, but their individual and combined predictive capabilities require further investigation.

Purpose of the Study:

• To evaluate the best-case predictive performance of individual and combined fall risk metrics.
• To assess the accuracy of Markov chain calculations for fall risk metrics compared to simulations.
• To develop and validate quadratic fall prediction models using Markov chain data.

Main Methods:

• Utilized a planar six-link bipedal model with curved feet, simulating walking at speeds from 0.8 to 1.2 m/s.
• Employed Markov chains to model gaits and estimate fall risk metrics, validating against brute force simulations.
• Developed and tested quadratic prediction models incorporating multiple fall risk metrics.

Main Results:

• Markov chains accurately calculated most fall risk metrics, with exceptions for short-term Lyapunov exponents.
• Individually, none of the 49 tested fall risk metrics accurately predicted the number of steps to fall.
• A combined model, excluding Lyapunov exponents, demonstrated substantially increased prediction accuracy.
• Prediction accuracy and precision improved with an increased number of steps used for metric calculation.
• 300-step simulations offered an optimal balance between accuracy and computational efficiency.

Conclusions:

• Effective fall prediction requires the integration of multiple fall risk metrics, rather than relying on single measures.
• The proposed Markov chain approach provides a computationally efficient method for estimating fall risk metrics.
• Combined metrics offer a more reliable measure of dynamic stability for fall prevention in humans and robots.