This paper examines recent research trends in which foundational models, including large-scale language models (LLMs) and vision-language models (VLMs), have enabled novel approaches to robotic autonomy and human-robot interfaces. Specifically, we focus on how vision-language-action models (VLAs) and large-scale behavior models (LBMs) contribute to enhancing the proficiency and functionality of robotic systems, and we review research moving toward agent-based applications and architectures. These studies range from exploring GPT-style tool interfaces to more complex systems in which AI agents act as coordinators, planners, cognitive agents, or general interfaces. These agent architectures enable robots to understand natural language commands, invoke APIs, plan task sequences, and support operations and diagnostics. Reflecting the rapidly evolving nature of this field, we cover not only peer-reviewed research but also community-driven projects, ROS packages, and industry frameworks. We propose a taxonomy for categorizing model integration approaches and provide a comparative analysis of the role agents play in various solutions across the current literature.
