Sonya Yang is a Software Development Engineer under AMD Research and Advanced Development (RAD). She oversees academic and research collaborations across the Asia-Pacific regions and focuses on AI models and sim-to-real research. Her work focuses on building simulation training workflows that enable robots to perform tasks more efficiently and safely, leveraging advanced algorithms to bridge the gap between virtual and real-world performance.

This talk offers an overview of AMD’s AI ecosystem, highlighting both software and hardware advancements. Key topics include ROCm 7, AMD’s latest open-source AI platform, and the Ryzen AI processors, which utilize NPU and iGPU acceleration for efficient on-device AI inference. We will delve into ROCm 7's newest features and discuss AMD's core focus areas in artificial intelligence. Additionally, the talk demonstrates the practicality of deploying robotics applications on AMD AI PCs. Participants will learn how to harness NPU computing power for essential computer vision tasks like face detection and depth estimation, and how to utilize open-source physical simulation tools within the AMD software environment. The session will also include demonstrations showcasing how to connect an AMD AI PC to a robotic arm, allowing attendees to experience firsthand the intuitive operation of robotic systems.

Wei-Fang Sun is currently a Solution Architect at NVIDIA AI Technology Center (NVAITC), where his main research areas include Reinforcement Learning, Generative Modeling, and Robotics applications and research.

The next wave of AI breakthroughs is expected to emerge from the widespread adoption of AI-powered robotics. By transforming industries such as manufacturing, healthcare, and transportation, intelligent robots are set to catalyze innovation and streamline operations across numerous sectors. This introductory course provides a concise overview of NVIDIA's robotics platform, designed to accelerate robotics development in this rapidly evolving field. Attendees will learn about leveraging NVIDIA Omniverse alongside OpenUSD to create high-fidelity digital twins for synthetic data generation and factory optimization. They will also be introduced to Isaac Sim, an advanced simulation tool that enables developers to build, test, and validate robotic applications in realistic, physics-accurate virtual environments. The session will also explore Isaac Lab, highlighting its role in training AI models through reinforcement learning (RL) and imitation learning (IL), enabling robots to perform complex tasks that traditional controllers struggle with. In addition, attendees will learn how to deploy robust robotic solutions using Isaac ROS, ROS2, and NVIDIA Jetson AGX, the world's most powerful embedded AI platform for robotics. By the end of the session, attendees will have a high-level understanding of how NVIDIA's ecosystem supports robotics innovation, and how they can leverage these software and hardware tools for their own future projects.

Shao-Hua Sun is an Assistant Professor in the Department of Electrical Engineering at National Taiwan University (NTU). He received his Ph.D. from the University of Southern California and his B.S. from NTU. His research spans machine learning, robot learning, reinforcement learning, and program synthesis. His work has been published at leading venues, including NeurIPS, ICML, CoRL, ICLR, etc. He has organized tutorials and workshops at NeurIPS, ICML, RLC, and CoRL.

Recent advances in artificial intelligence and machine learning have dramatically enhanced machines’ abilities to perceive visual scenes, understand natural language, and even surpass human experts in complex games. Yet, enabling robots to operate robustly in unstructured environments, manipulate novel objects, and acquire new skills—ultimately to relieve humans from tedious or hazardous tasks—remains a formidable challenge. My research aims to build a program-guided robot learning framework that is both interpretable and generalizable, allowing robots to acquire long-horizon, hierarchically structured skills such as furniture assembly and cooking. In this framework, desired behaviors are represented as executable programs, while primitive skills are learned as reusable modules for carrying out higher-level plans. This talk will present a series of projects that progressively develop and integrate these ideas, advancing toward a unified framework for interpretable, generalizable, and skillful robot learning.