Physical World

Research on the physical world within the context of artificial intelligence focuses on understanding and interacting with the physical environment through embodied AI systems, often leveraging large language models (LLMs) and deep neural networks (DNNs). Current research emphasizes the vulnerabilities of these systems to adversarial attacks, including those manifested through physical objects or manipulations of sensor inputs (e.g., thermal or visible light), and explores methods to improve robustness and safety. This work has significant implications for the security and reliability of AI systems in real-world applications, such as autonomous vehicles and robotics, as well as for advancing our understanding of how AI interacts with and interprets physical phenomena.