Program

Abstract

Generative Physical AI in robotics represents a transformative approach that fuses advanced generative models from artificial intelligence with the dynamic world of physical robotic systems. At its core, generative physical AI involves the creation of models that can simulate, predict, and optimize robotic behaviors in real-time. Unlike traditional control algorithms, these generative models are designed to learn and generate possible system states and responses based on a wide array of sensory inputs and environmental interactions. This capability allows robots to not only adapt to unexpected changes in their surroundings but also to innovate solutions by autonomously exploring diverse potential actions. A key element of this paradigm is the integration of knowledge-aware techniques. By incorporating structured knowledge, generative physical AI frameworks can tap into rich, interconnected datasets that provide contextual understanding and enhance decision-making. These structured representations allow robotic systems to map complex relationships, reason over vast repositories of domain-specific information, and ground their learning processes in a broader context. Consequently, this dual approach ensures that robotic systems are better equipped to navigate complex scenarios, understand nuanced environments, and optimize their operations accordingly.

This paradigm leverages machine learning techniques such as deep generative models, probabilistic reasoning, and reinforcement learning to create a seamless bridge between simulated predictions and tangible implementations. By continuously refining models with real-world feedback, generative physical AI systems can iteratively improve their performance, enhance their adaptability, and foster robust decision-making in complex environments. This approach empowers robots to perform tasks with a level of sophistication that mirrors human intuition, making them more effective in dynamic scenarios like disaster response, autonomous navigation, and intricate manufacturing processes.

In this workshop, we will dive into the topic “Generative Physical AI in robotics” with invited speakers, a poster session, and a round table for active discussion with the participants. Academic and industry speakers will be involved to provide insights at different levels, highlighting current developments, limitations, needs/requirements, and challenges. Real-world applications (in the field of autonomous driving, industrial robotics, healthcare, etc.) will be discussed to underline current and future directions.