Towards Comprehensive Multimodal Perception: Introducing the Touch-Language-Vision Dataset