Embodied AI is rapidly moving from laboratory research toward real-world applications. With the rapid advancement of Vision-Language-Action (VLA) models, robots are evolving from “understanding the world” to “acting in the world.” At the same time, building open, trustworthy, and reproducible evaluation systems, and leveraging open collaboration to accelerate technological progress, have become critical challenges for the development of embodied AI.

This keynote will present insights from the RoboChallenge 2025 Annual Report, sharing the latest evaluation results of global embodied AI foundation models in real-world robotic environments. It will also highlight RoboChallenge’s latest progress, including real-robot challenges at leading international conferences such as CVPR and ICRA, as well as the launch of EAI Bench at the AI for Good Global Summit in Geneva.

Based on tens of thousands of real-robot execution trials, the report analyzes the current state and future trends of embodied AI models. Current leading Vision-Language-Action (VLA) models have achieved approximately 62% success rate on the Table30 benchmark, while success rates for complex fine-manipulation tasks remain below 15%, demonstrating that embodied AI is still at a critical stage of advancing from semantic understanding toward robust physical interaction and complex manipulation capabilities. Meanwhile, RoboChallenge has attracted developers from China, the United States, Singapore, Japan, and other regions, with open-source models, open datasets, and open evaluation becoming increasingly important drivers of innovation.

In the era of embodied AI, open source is evolving from software sharing toward broader open collaboration. The future requires collective efforts to advance open models, datasets, robot interfaces, evaluation benchmarks, and open standards. RoboChallenge aims to connect global developers, robot manufacturers, research institutions, and industry partners to build an open, trustworthy, and reproducible real-robot evaluation ecosystem, enabling more developers to access real-world robotic resources and accelerate embodied AI innovation.

Speakers:

Emily Chen: Co-Founder of Kaiyuanshe

Emily Chen is the Co-founder of Kaiyuanshe and a current Board Director . A long-time advocate, builder, and connector in the global open source community, she has played a key role in shaping open collaboration across Asia and globle.

Emily served on the GNOME Foundation Board of Directors in 2010 and founded the GNOME.Asia community in 2008, bringing the GNOME.Asia Summit to over ten countries and regions. She was honored with the GNOME Foundation’s highest recognition, the GNOME Pants Award, at GUADEC 2014 in Sweden.

From 2017 to 2025, she hosted multiple visits of GitHub’s executive team and global community leaders to China, building bridges between international platforms and the Chinese open source ecosystem. She also served as the China representative to the Open Source Initiative (OSI) Alliance and was Deputy Secretary-General of the China OSS Promotion Union in 2016.

Emily initiated impactful projects such as the Kaiyuanshe China OPenSource Annual Report, COSCon (China Open Source Conference), and the “33 Open Source Pioneers” list. She is a core contributor to Mozilla, a Google Summer of Code mentor, and one of the organizers of the Open Source Congress 2024. Currently, she serves as a TOC Mentor at the OpenAtom Open Source Foundation.

With nearly two decades of experience in open source community building and global-local collaboration, Emily continues to champion open innovation in the GenAI era.