Indian-Population-Pose-Asymmetry-Dataset-v1

About Dataset

The Indian Population Pose Asymmetry Dataset (IPPAD v1.0) contains 3,000 samples of bilateral biomechanical features extracted from simulated 17-keypoint human skeletal poses. Each sample encodes 22 features including bilateral joint angles, limb length asymmetry indices, shoulder and hip height differences, trunk lean angle, and anthropometric ratios. The dataset is calibrated to Indian population norms sourced from the SIZE INDIA anthropometric reference (mean height 163 cm, SD 8 cm), making it one of the few robotics-oriented biomechanical datasets grounded in Indian population statistics. Samples are classified into three posture asymmetry categories: Symmetric (class 0), Mild Asymmetry (class 1), and Severe Asymmetry (class 2), with 1,000 samples per class for a balanced distribution. Asymmetry is modelled via controlled lateral keypoint displacement on one side of the skeleton, simulating real-world conditions such as scoliosis, injury-related compensatory postures, and habitual asymmetric loading common in Indian industrial and agricultural workers. The dataset was used to train a soft-voting ensemble classifier (Random Forest + Gradient Boosting) achieving 90% test accuracy and macro F1 of 0.90 across all three classes. The dataset is intended as a benchmark resource for human-robot interaction research, ergonomics screening tools, and assistive robotics systems that must adapt robot trajectories based on the physical asymmetry of a human subject.