Learning-Based Motion-Intention Prediction for End-Point Control of Upper-Limb-Assistive Robots.

Area of Science:

• Robotics
• Human-Robot Interaction
• Biomedical Engineering

Background:

• Upper-limb assistive devices often lack intuitive control, hindering their practical application.
• Effective human-robot interaction is crucial for seamless operation of assistive technologies.

Purpose of the Study:

• To develop a novel learning-based controller for assistive robots that intuitively predicts desired end-point positions using onset motion.
• To evaluate the efficacy of multi-modal sensing (IMUs, EMG, MMG) for motion intention detection.

Main Methods:

• Implemented a multi-modal sensing system with IMUs, EMG, and MMG sensors.
• Collected kinematic and physiological data during reaching and placing tasks from five healthy subjects.
• Utilized traditional regression and deep learning models (including RNNs) to predict hand position from onset motion data.

Main Results:

• Inertial Measurement Unit (IMU) data alone proved sufficient for accurate motion intention detection, comparable to using additional EMG or MMG sensors.
• Recurrent Neural Network (RNN) models demonstrated effectiveness in predicting target positions within short onset windows for reaching and longer horizons for placing tasks.

Conclusions:

• The proposed prediction model using IMUs offers a viable and efficient method for motion intention detection in assistive robots.
• This approach can significantly enhance the usability and intuitiveness of upper-limb assistive and rehabilitation robots.