Exploring the Adversarial Vulnerabilities of Vision-Language-Action Models in Robotics