A Comprehensive Multimodal Gait Dataset for Lower-Limb Wearable Robotics in Indoor and Outdoor Environments

Abstract

Daily walking locomotion exhibits variability, asymmetry and ground constraints. Recording and analyzing real-world locomotion data is essential for expanding the application scenarios of wearable robots such as exoskeletons and prostheses. However, most existing gait datasets are collected with specialized equipment under predefined laboratory conditions, lacking visual information from natural walking environments and annotations for transitional states across different surfaces. To address this problem, we collected walking data from 17 healthy participants across diverse indoor and outdoor environments, including kinematic data from seven inertial measurement units (IMUs), foot pressure data from force sensing resistors (FSRs) with 96 points, and visual data from three front-facing cameras with 1280 × 720 resolution. All modalities are time-synchronized, with 448 clips and 243607 data frames in total. Full annotations are provided, covering 16 classes of lower-limb locomotion states (8 steady states, 8 transitional states), bilateral gait phases and bilateral coordinates of the center of pressure (COP). This dataset can be utilized for exploring the applications of machine learning in locomotion state recognition and intelligent control of lower-limb wearable robotics.