Offline and Real-Time Implementation of a Terrain Classification Pipeline for Pushrim-Activated Power-Assisted...

Area of Science:

• Assistive technology
• Robotics
• Machine learning

Background:

• Pushrim-activated power-assisted wheelchairs (PAPAWs) enhance mobility for users.
• Adapting PAPAW control to different terrains is crucial for usability.
• Terrain classification is key for developing adaptive PAPAW controllers.

Purpose of the Study:

• To investigate the impact of model parameters on terrain classification accuracy for PAPAWs.
• To evaluate the effectiveness of learning-based models for real-time PAPAW control.
• To identify optimal configurations for accurate and efficient terrain classification.

Main Methods:

• Examined the influence of feature characteristics, terrain types, and data segment length on classification accuracy.
• Utilized Random Forest classifiers for offline and real-time terrain classification.
• Compared performance based on various parameter combinations and data processing techniques.

Main Results:

• Random Forest classifiers demonstrated computational efficiency for real-time applications.
• Highest accuracy was achieved using a combination of time- and frequency-domain features.
• Increased data segment length and grouped similar terrains improved classification accuracy.

Conclusions:

• Real-time terrain classification using Random Forest is feasible for PAPAWs.
• Optimized feature selection and data processing enhance classification performance.
• Findings support the development of adaptive PAPAW controllers for diverse terrains.