Deep Learning Architecture to Assist With Steering a Powered Wheelchair

Area of Science:

• Robotics
• Artificial Intelligence
• Biomedical Engineering

Background:

• Powered wheelchairs are crucial for mobility but navigating complex environments poses challenges.
• Existing systems may lack sophisticated obstacle avoidance and intuitive control.
• The integration of advanced AI for assistive devices is an emerging area of research.

Purpose of the Study:

• To develop and evaluate a novel deep learning architecture for powered wheelchair steering assistance.
• To enhance wheelchair safety and usability for disabled drivers.
• To explore the application of Long Short Term Memory (LSTM) neural networks in assistive navigation.

Main Methods:

• A Long Short Term Memory (LSTM) neural network was trained using a rule-based approach.
• Input data included joystick commands (speed, direction) and ultrasonic transducer readings.
• The system generated a safe steering direction by blending user input with obstacle avoidance data.

Main Results:

• The deep learning model successfully predicted safe wheelchair directions by integrating user intent and environmental perception.
• Obstacle avoidance was achieved by combining desired trajectories with collision-prevention maneuvers.
• The system demonstrated the feasibility of using LSTMs for real-time steering assistance.

Conclusions:

• The novel deep learning architecture offers a promising approach to enhance powered wheelchair steering.
• LSTM networks can effectively process multi-modal inputs for intelligent assistive navigation.
• The system provides a safer and more intuitive user experience, with the option for manual override.