Building the Open Embodied AI Ecosystem:Infrastructure, Data, Models, and Paths to Commercialization
Embodied AI is bringing artificial intelligence from the digital world into the physical world. Yet its development continues to face significant challenges, including closed software ecosystems, high-cost data, and fragmented toolchains. This keynote will explore how to build an open, portable, and commercially sustainable embodied AI ecosystem across three foundational layers: infrastructure, data, and models.
Drawing on Motphys’s experience in developing simulation and training platforms, open simulation benchmarks and evaluation frameworks, as well as leading open academic collaborations such as GS-Playground and UniLab, the talk will examine how open co-development and cross-platform collaboration can reduce dependence on any single hardware architecture or proprietary software ecosystem. The goal is to enable simulation, training, and inference workloads to run more flexibly across platforms from NVIDIA, AMD, Intel, and a broader range of European and Chinese computing hardware providers.
The keynote will also discuss how simulation infrastructure can serve as the foundation for integrating internet-scale video, human demonstrations, real-robot data, and synthetic simulation data into a multimodal data hub for embodied AI training. Such an infrastructure can support the continuous accumulation, reuse, and closed-loop iteration of data across different tasks, models, and robotic embodiments.
As generative AI reshapes the software industry and challenges traditional open-source models, the talk will revisit the boundaries of openness: which foundational capabilities should be openly shared, and which data assets, models, and industry-specific capabilities should remain sources of commercial differentiation. Open source must establish a new balance among technological autonomy, ecosystem collaboration, commercial returns, and long-term maintenance.
Scaling embodied AI requires an open ecosystem that connects infrastructure, data, models, computing hardware providers, robotics companies, research institutions, and developers.
Speakers:
Jiaqi Sun: Motphys, Co-founder (谋先飞联合创始人)
Jiaqi Sun, co-founder of Motphys, PhD from the Department of Mechanics at Peking University. He has extensive research and industry experience in physics simulation, computational geometry and graphics algorithms, heterogeneous CPU/GPU computing, and the integration of artificial intelligence with scientific computing. For more than a decade, he has worked at the forefront of industrial physics solver development, leading the creation of integrated software that connects geometric processing with multiphysics simulation. The resulting technologies and products have been applied across automotive engineering, aerospace, data centers, renewable energy, semiconductor manufacturing, and other industrial simulation domains.
In recent years, Jiaqi has focused on integrating physics simulation with embodied AI at Motphys. He has driven the development of the Motrix family of simulation and training platforms and the MotrixBench open simulation benchmarking framework, while also supporting academic collaborations such as GS-Playground and UniLab. His work explores open infrastructure for Real2Sim2Real pipelines, robot learning, and heterogeneous computing platforms. He also studies how to build an open-source embodied AI ecosystem that balances technological independence, collaborative innovation, and long-term commercial sustainability.
