At the 2026 World Robot Congress (WRC), Volcano Engine held its Embodied AI Technology and Applications Forum, unveiling its latest technological advancements, product roadmap, and industry implementation outcomes in the embodied intelligence sector.
During the event, Yang Liwei, Vice President of Volcano Engine and Dean of the Embodied Intelligence Engineering Institute, analyzed current industry trends. He noted that capital enthusiasm in the embodied intelligence sector continues to heat up, industry model development paths are becoming increasingly clear, and favorable conditions—including declining hardware costs and expanding real-world application scenarios—are accelerating industry deployment. He also highlighted that engineering expertise from large language models is gradually transferring to the embodied intelligence domain, with technologies like In-Context Learning enabling rapid response to demands through minimal sample examples, offering valuable references for optimizing embodied intelligence tasks.
Yang Liwei also detailed key progress in Volcano Engine's foundational embodied intelligence models. First, the team launched the full-duplex speech model Seeduplex, designed to address the high real-time and multi-information fusion interaction requirements of embodied scenarios. This model overcomes the shortcomings of traditional model-assembly approaches, such as slow response times and low information transmission efficiency, by integrating visual and speech inputs to perceive environmental changes in real time and proactively adapt to user interaction needs.
Second, the Seed2.1 multimodal model achieves integrated perception, understanding, and execution capabilities. Currently, over half of China's embodied intelligence companies and multiple autonomous driving firms rely on this model's visual-language understanding capabilities for data annotation tasks.
On the ecosystem collaboration front, Yang Liwei revealed that 15 of China's top 20 highest-valued embodied intelligence companies have established deep partnerships with Volcano Engine. By coordinating efforts between its foundational model team and engineering institute, Volcano Engine is breaking down barriers to scenario deployment, leveraging real-world engineering practices to drive a closed-loop data flywheel, and continuously feeding back into foundational model optimization. The company also publicly opened industry collaboration and talent recruitment channels.