Global Simulation Conference 2026

Conference Preview | Featured Speaker Line

Scan Here!

November 13–16 · Hangzhou, China

AI-Driven Global Simulation Innovation and Collaborative Governance

Main Forum · 10 Thematic Sessions · International Simulation Technology Exhibition · Business Matchmaking

As the Global Simulation Conference 2026 approaches, more speakers for the thematic forums are being announced. Today, we introduce two speakers from the Physical AI, Intelligent Simulation, and Data Governance Forum: HE Ye, Senior Software Architect at AMD’s AI Business Unit, and SUN Jiaqi, Co-founder and Chief Architect of Motphys and a PhD in Computational Mechanics from Peking University. Together, they will explore the full pipeline “From Physics-Based Simulation to Robot Training”, focusing on Motphys’ open-source UniLab project and the AMD ROCm ecosystem. Their presentation will show how the self-developed MotrixSim physics engine can bridge high-precision industrial physics simulation and large-scale robot simulation, while native support for AMD ROCm enables 3–10× end-to-end acceleration through heterogeneous computing architectures. They will also introduce an open-source infrastructure for embodied AI based on the Real2Sim2Real approach.

HE Ye, Senior Software Architect in AMD’s AI Business Unit

He Ye is a Senior Software Architect in AMD's AI Business Unit, focusing on technology expansion and solution enablement for Physical AI on the AMD ROCm ecosystem. He is also a contributor to open-source projects such as UniLab/MortrixLab. Working closely with MOU Xianfei, he helped bring native ROCm support to UniLab/MortrixLab, making them a key open-source solution for robot simulation and reinforcement learning on AMD ROCm.

SUN Jiaqi, Co-founder and Chief Architect of Motphys   

SUN Jiaqi is Co-founder and Chief Architect of Motphys, holding a PhD in Computational Mechanics from Peking University. He has long been engaged in research and industrial practice spanning physics simulation, geometry and graphics algorithms, CPU/GPU heterogeneous parallel computing, and the integration of AI with scientific computing. For over a decade, he has worked at the frontier of industrial-grade physics solver development, with simulation algorithm experience covering fluid dynamics, conjugate heat transfer, fluid-structure interaction, and nonlinear structural analysis (large deformation, material nonlinearity, contact) across multiple physics domains. He has led the development of integrated geometry processing and multiphysics solving software, and the related technologies and products have been applied in industrial physics simulation fields such as automotive, aerospace, data centers, new energy, and semiconductors.

In recent years, at Motphys he has been dedicated to integrating physics simulation with embodied intelligence, driving the development of the Motrix series simulation training platforms and the MotrixBench open simulation benchmarking system, and supporting academic collaboration projects. His work is building open infrastructure for the Real2Sim2Real pipeline, robot learning, and diverse heterogeneous computing platforms, while also exploring how to build an open-source embodied intelligence ecosystem that balances technological autonomy, collaborative innovation, and long-term commercial sustainability.

Conference Preview

Speakers: HE Ye, SUN Jiaqi

Topic: From Physics Simulation to Robot Training: Deploying UniLab RL with ROCm

Session: Physical AI, Intelligent Simulation, and Data Governance Thematic Forum

 Session Preview

Embodied intelligence moving into the real world requires simulation infrastructure that supports physical interaction, policy learning, and capability validation. Taking Motphys's open-source project UniLab as its main thread, this report presents the complete pipeline “from physics solving to robot training”: the self-developed MotrixSim physics engine bridges high-precision industrial physics solving with large-scale robot simulation; a heterogeneous architecture that decouples CPU simulation from GPU learning breaks away from the GPU-resident paradigm; and with UniLab's native support for AMD ROCm—spanning AMD APUs (Strix Halo / Strix Point), Radeon GPUs, and Instinct GPUs—it builds a robot reinforcement learning platform for diverse GPUs from the edge to the data center.

The report covers three aspects. First, for contact, friction, and fine-manipulation scenarios, it discusses the trade-offs among physical solving accuracy, numerical stability, and computational efficiency, and why contact-rich tasks favor CPU physics backends. Second, it introduces UniLab's heterogeneous pipeline of CPU rollout workers, a lock-free shared-memory runtime, a GPU learner, and asynchronous weight synchronization, along with the scheduling of PPO, APPO, FastSAC, and FlashSAC on the same runtime. From the ROCm perspective, it shares porting experience and first-hand benchmarks showing 3–10× end-to-end speedup over GPU-centric baselines on the same hardware, with AMD platforms compared against an RTX 4090. Third, it demonstrates sim-to-real validation on real robot tasks and explores a Real2Sim2Real loop in which real data identifies physical parameters and calibrates simulation assets.

Finally, the report explores the use of generative and multimodal models in asset production and training automation, and discusses—drawing on open-source practice—how open interfaces and shared benchmarks can reduce toolchain integration costs, balancing technological autonomy, ecosystem co-construction, and commercial sustainability to provide robot companies and developers with sustainably evolving simulation infrastructure.

The Global Simulation Conference 2026 brings together leading experts from around the world to explore cutting-edge topics, including Physical AI, intelligent simulation and data governance, embodied intelligence, and industrial foundation models. The conference will feature a main forum and 10 thematic forums covering key areas such as aerospace, smart healthcare, low-altitude transportation, green and intelligent marine equipment, and youth innovation and development.

YE He and Jiaqi Sun will bring an end-to-end perspective spanning industrial-grade physics simulation, computing hardware, and the open-source ecosystem for embodied AI. We look forward to exploring with you the latest developments and applications of simulation technology in Physical AI, industry, and open-source collaboration.

For Registration / Exhibition Inquiries: Mr. YANG Tel: 18501232229 

Email: cstia@ccsat.fun

Scan to Register

GloSim 2026

From “Simulation for Everything” to “AI-Driven Global Collaboration”

Simulation technology is redefining the boundaries of innovation.

See You in Hangzhou · November 2026

 Conference Schedule

Global Simulation Conference 2026 Schedule