An Efficient Wi-Fi Sensing Method for Robotic Arm Motion Recognition.

Area of Science:

• Robotics
• Wireless Communications
• Machine Learning

Background:

• Channel State Information (CSI)-based sensing offers a contactless, low-cost method for understanding robotic arm motion.
• Existing CSI-based methods struggle with weak, localized CSI perturbations and identifying key temporal information in robotic motion sequences.

Purpose of the Study:

• To propose an efficient multi-stage method, MSPoolNet, for accurate robotic arm motion recognition using CSI.
• To address challenges in fine-grained feature extraction and adaptive temporal selection for robotic motion sensing.

Main Methods:

• Developed MSPoolNet with three modules: adaptive temporal downsampling for local pattern extraction, temporal gating for adaptive feature reweighting, and a Transformer-based encoder using pooling for efficient global interaction.
• Processed raw CSI signals to extract local patterns and dynamically highlighted crucial temporal segments.

Main Results:

• MSPoolNet achieved state-of-the-art performance on two public datasets, exceeding 99% accuracy.
• The method demonstrated effectiveness in handling weak CSI perturbations and sparse discriminative information.
• Maintained a compact model size while delivering high accuracy.

Conclusions:

• MSPoolNet offers an efficient and accurate solution for robotic arm motion recognition using CSI.
• The proposed modules effectively address the limitations of existing CSI-based sensing methods for robotic applications.