A Probability Fusion Approach for Foot Placement Prediction in Complex Terrains.

Area of Science:

• Biomechanics and Robotics
• Assistive Technology
• Machine Learning for Human Motion Analysis

Background:

• Existing foot placement prediction methods are limited to level ground walking.
• Complex terrains (ramps, stairs, obstacles) pose significant challenges for current assistive walking technologies.
• Accurate foot placement prediction is crucial for enhancing mobility and safety in daily life for individuals with lower-limb disabilities.

Purpose of the Study:

• To develop and validate a robust foot placement prediction method for complex terrains.
• To improve assistive walking systems for individuals with lower-limb impairments in diverse environments.
• To enable more natural and safe navigation over uneven and varied ground surfaces.

Main Methods:

• A probability fusion approach combining deep learning and environmental constraints.
• Training a deep learning model on augmented data to predict preliminary foot placement probability distribution.
• Calculating a feasible landing area using environmental information and human walking constraints.
• Fusing the probability distribution with the feasible area for final foot placement prediction upon heel-off detection.

Main Results:

• Achieved a root mean square error (RMSE) of 8.19 ± 1.20 cm, representing less than 8% of average stride length.
• Demonstrated a landing feasible area accuracy (LFAA) of 95.11 ± 3.09% in experiments.
• The proposed method showed faster and more accurate predictions compared to existing studies in complex terrains.

Conclusions:

• The probability fusion approach effectively predicts foot placement in complex terrains.
• This advancement holds significant potential for improving assistive devices for individuals with lower-limb disabilities.
• The method offers a more reliable solution for real-world walking assistance in challenging environments.