Self-Initialized Locomotion Mode Prediction with GPU-Free Terrain Reconstruction.

Area of Science:

• Robotics
• Human-Computer Interaction
• Biomechanics

Background:

• Accurate Locomotion Mode Prediction (LMP) is crucial for lower-limb wearable robots to adapt assistance across varied terrains.
• Existing reconstruction-based LMP methods offer high accuracy but suffer from significant computational costs (GPU dependency) and require manual initialization.
• These limitations hinder the practical deployment of advanced LMP systems in real-world scenarios.

Purpose of the Study:

• To develop a self-initialized, GPU-free Locomotion Mode Prediction (LMP) method for lower-limb wearable robots.
• To overcome the deployment constraints of existing LMP techniques, specifically high computational cost and manual initialization.
• To enable reconstruction-based LMP entirely on an onboard CPU without wearer or professional supervision.

Main Methods:

• Proposed a novel method utilizing a gravity-aligned world coordinate frame as a unified geometric reference.
• Implemented a self-initialization procedure to establish this reference frame autonomously.
• Developed a progressive plane representation for GPU-free terrain reconstruction, integrated into a complete LMP pipeline.

Main Results:

• The proposed method achieved prediction accuracy comparable to state-of-the-art, GPU-dependent approaches.
• Demonstrated successful operation entirely on an onboard CPU, eliminating the need for GPU acceleration.
• Achieved high initialization success rates and efficient computational performance across diverse terrains and subjects.

Conclusions:

• The developed self-initialized, GPU-free LMP method effectively addresses the deployment challenges of wearable robotic systems.
• This approach enables accurate and adaptive robotic assistance without manual intervention or specialized hardware, paving the way for wider adoption.
• The findings highlight the potential of CPU-based, autonomous terrain reconstruction for real-time robotic applications.