TIC-VLA: A Think-in-Control Vision-Language-Action Model for Robot Navigation in Dynamic Environments