Learning-Aided User Intent Estimation for Smart Rollators.

Area of Science:

• Robotics
• Human-Computer Interaction
• Biomedical Engineering

Background:

• The aging population and increased mobility disabilities necessitate advanced assistive devices.
• Smart rollators require sophisticated control systems for enhanced safety and reliability.
• Accurate prediction of user intent is crucial for intuitive and responsive rollator control.

Purpose of the Study:

• To develop and validate a classification method for predicting rollator user intent.
• To utilize indirect sensor inputs for intent estimation.
• To improve the safety and usability of smart rollator technology.

Main Methods:

• Implemented an intent prediction system using an inertial measurement unit (IMU) and an encoder on a modified robotic rollator platform.
• Employed the K-Nearest Neighbors (KNN) classification algorithm to analyze sensor data.
• Collected data from 7 healthy young adult participants performing various movement tasks.

Main Results:

• The KNN classification algorithm achieved 92.9% accuracy in predicting three distinct user intents: turn left, turn right, and walk straight.
• The developed intent estimation method was experimentally verified on the robotic platform.
• Indirect sensor inputs proved effective for rollator user intent prediction.

Conclusions:

• The proposed classification method effectively predicts rollator user intent with high accuracy.
• This technology has the potential to significantly enhance the safety and functionality of smart rollators.
• Further research can explore real-world applications and diverse user populations for this intent prediction system.