Parkinson's disease is a neurodegenerative disorder, characterized by progressive motor impairments including tremor, rigidity, and bradykinesia. Clinical assessment relies primarily on the Movement Disorder Society-Unified Parkinson's Disease Rating Scale (MDS-UPDRS), which, despite being the gold standard, suffers from inherent subjectivity, inter-rater variability, and limited sensitivity to subtle motor changes, particularly for distinguishing between mild severity levels (scores 0-2). These limitations underscore the need for objective, quantitative assessment methods that can complement clinical evaluation. Recent advances in skeleton-based motion encoder networks have demonstrated remarkable capabilities in capturing complex human movement patterns from video or motion capture data. However, their application to clinical gait assessment in Parkinson's disease faces significant challenges: limited availability of labeled clinical data, underrepresentation of pathological movement patterns in training datasets, and substantial cross-dataset generalization gaps arising from variations in recording protocols and sensor modalities.