WearMoCap: multimodal pose tracking for ubiquitous robot control using a smartwatch.

Area of Science:

• Robotics
• Human-Computer Interaction
• Wearable Technology

Background:

• Human pose tracking is crucial for robotics and human-computer interaction.
• Existing motion capture systems are often cumbersome and not ubiquitous.
• Smartwatch sensors offer a potential for unobtrusive and accessible motion tracking.

Purpose of the Study:

• To introduce WearMoCap, an open-source library for human pose tracking using smartwatch sensor data.
• To enable ubiquitous robot control through accurate pose predictions.
• To evaluate different sensor configurations for motion capture.

Main Methods:

• Developed WearMoCap with three modes: Watch Only, Upper Arm, and Pocket.
• Collected large-scale datasets from 8 human subjects using consumer-grade devices.
• Evaluated performance in handover and teleoperation tasks, comparing against a gold-standard motion capture system.

Main Results:

• WearMoCap achieved accuracy within 2 cm of gold-standard motion capture systems.
• The Upper Arm mode yielded the most accurate wrist position estimates (6.79 cm RMSE).
• Demonstrated real-robot application performance in handover and teleoperation tasks.

Conclusions:

• WearMoCap provides a robust and accurate solution for smartwatch-based human pose tracking.
• The open-source library facilitates future research in wearable motion capture for robotics.
• The system is designed for ubiquitous applications where motion tracking is essential.