Can Vision-Language Models Understand and Interpret Dynamic Gestures from Pedestrians? Preliminary Datasets and Exploration for Instructive Nonverbal Commands in Cooperative Autonomous Vehicles