Multi-Modal Graph Convolutional Network with Sinusoidal Encoding for Robust Human Action Segmentation