Towards the Development of a Learning-Based Intention Classification Framework for Pushrim-Activated Power-Assisted...

Area of Science:

• Rehabilitation Engineering
• Human-Computer Interaction
• Robotics

Background:

• Manual wheelchairs (MWCs) require significant physical exertion.
• Power assist devices, like pushrim-activated power-assisted wheels (PAPAWs), aim to reduce this load.
• Current PAPAW controllers may not accurately interpret user intentions, leading to maneuvering difficulties.

Purpose of the Study:

• To analyze wheelchair propulsion dynamics with manual versus powered wheels.
• To design and evaluate a user-intention detection framework for PAPAWs.
• To improve the control and user experience of powered assist devices for manual wheelchairs.

Main Methods:

• Calculated wheelchair linear and angular velocity from wheel angular velocity data.
• Collected kinematic data from manual wheelchair experiments.
• Tested six supervised learning algorithms for classifying user movements (not moving, straight, left turn, right turn).

Main Results:

• Identified suboptimal design in current powered wheel controllers.
• All tested supervised learning algorithms achieved high accuracy in detecting wheelchair movement types.
• Classification of movements was completed with low computational time.

Conclusions:

• The proposed user-intention detection framework accurately classifies MWC movements.
• This framework can inform the development of adaptive, learning-based controllers for PAPAWs.
• Implementing individualized control strategies can enhance the usability and experience of PAPAW users.