Industrial robots should possess the capability to discern human intentions by analyzing partially observed human actions and subsequently carry out assisting tasks. Nevertheless, conventional HRC in both industrial and academic settings concentrate on either reactive or adaptive robot planning. Such approaches involve waiting for human actions, potentially resulting in unanticipated safety concerns due to delayed responses. This impedes the seamless transition of HRC toward predictable teamwork. To address this bottleneck, we introduce an approach that leverages multimodal transfer learning to enable ongoing prediction of human intentions and proactive decision-making by robots. This approach aims to foster predictability in collaborative efforts in the near future. We evaluate the effectiveness of our method by demonstrating its application in an aircraft bracket assembly task.