During the 2026 World Robot Conference (WRC), a dedicated forum on embodied intelligence technology and applications was hosted by Volcano Engine on the afternoon of August 24th. At this event, Wang Yunkai, who leads the embodied intelligence product line at Volcano Engine, delivered a keynote speech outlining the company's strategic approach to empowering the embodied industry and the tangible results already achieved in this sector.
From his perspective, the primary bottleneck currently facing the embodied intelligence field has shifted from the sheer volume of data to the quality of that data. "EGO and UMI have certainly scaled up in volume, but the real question is whether the quality matches that increased quantity," he remarked. He pointed out that a million hours of raw dataset might contain the same amount of genuinely useful information as just a hundred thousand hours. To address this challenge, Volcano Engine has established a comprehensive toolchain that spans the entire data lifecycle.
According to his description, the Doubao large model is now being utilized by numerous leading companies to perform data slicing and fine-grained frame-level annotation, which supports their processing needs of hundreds of thousands of data points per month. The Seedance model can generate third-person perspective views and replace bare-hand images with robotic arms for data augmentation purposes. Meanwhile, SimReady leverages Seed 3D to create simulation assets and enriches them with semantic labels and physical properties, making them directly deployable in simulation environments.
On the interaction front, Doubao has already implemented an end-to-end S2S voice solution and is actively advancing research on full-modality models, with plans to explore an interactive and explorable world model. Wang Yunkai also disclosed that nearly all embodied robots featured on this year's Spring Festival Gala stage relied on the Doubao model for their interactive capabilities, as leading partner manufacturers consistently chose this technology.