UR 2026 is pleased to host the following tutorial. Please check back for updates on tutorial dates, times, and locations.
Tutorial #1: Physical AI: From Simulation to Cross-Domain Robot Learning on AI PCs
Organizers: Shan-Ya Yang* (Advanced Micro Devices, Inc); JIAHUA LU (AMD); Ping-Chun Hsieh (National Yang Ming Chiao Tung University); Ming-Hong Chen (National Yang Ming Chiao Tung University); You-De Huang (National Yang Ming Chiao Tung University)
Abstract: Physical AI is opening new opportunities for robotics, yet many current workflows remain fragmented across edge hardware, simulation, and robot learning pipelines. This tutorial presents an end-to-end and fully open-source workflow for developing Physical AI and robot learning workflow. The tutorial first introduces AI PCs as an accessible platform for robotics. By combining CPUs, iGPUs, and NPUs, AI PCs enable perception and robot policy to run close to the physical system. Participants will learn how such edge hardware can support robotics workloads including ROS-based pipelines, local LLM agents, and VLM/VLA-enabled applications. The tutorial then demonstrates how edge computing resources can be used to build a robotic simulation platform for Physical AI. Using the open-source simulation Platform Genesis with the open-source ROCm stack, participants will learn how simulation supports robot testing and training before deployment to real-world systems. Hands-on content includes importing a robotic arm into a virtual environment, performing inverse-kinematics motion control, and implementing parallelized simulation environments for scalable experimentation. Building on this foundation, the tutorial introduces reinforcement learning in simulation as a practical robotics workflow. It covers task design, policy training in digital twins, and behavior adaptation under changing conditions. A central focus is cross-domain learning and cross-robot transfer: how knowledge transfer and pre-trained policy adaptation can be achieved across robot arms with similar degrees of freedom and joint structures, even when the robot models differ. A key strength of this tutorial is its strong commitment to the open-source ecosystem. All demonstrations and hands-on materials will be released with full source code on GitHub, enabling attendees to reproduce the workflows for research and education. Overall, this tutorial provides a reproducible and scalable pathway for developing Physical AI on affordable edge platforms.
Keywords: World Modelling; Motion Planning and Obstacle Avoidance; Manipulation Planning and Control