Beijing startup DeepCybo raises $14M to train robots with first-person human data
What's the deal? DeepCyboDealroom has a profile for this one. Try Dealroom →, a Beijing-based startup focused on embodied artificial general intelligence (AGI), has raised approximately $14M (100 million yuan) in a recent funding round. The company is building what it calls "embodied world models" by collecting and training on first-person human activity data — essentially teaching robots to understand the physical world the way humans experience it.
The round was led by Sinovation Ventures, the VC firm founded by AI luminary Kai-Fu Lee. DeepCybo was founded in 2024 by a team with backgrounds in computer vision, robotics, and large language models.
Why now? The race to build robots that can operate intelligently in unstructured, real-world environments is heating up globally. Most approaches rely on simulation or third-person video data, but DeepCybo argues that first-person human data — captured from the perspective of someone performing a task — gives models a richer understanding of how to interact with physical spaces.
China's robotics sector has seen a surge of investment as the country pushes to lead in embodied AI, a field seen as the next frontier after the large language model boom. DeepCybo is positioning itself at the intersection of two hot trends: foundation models and humanoid robotics.
What could go wrong? Collecting large-scale, high-quality first-person human data is expensive and logistically complex. The company faces stiff competition from well-funded rivals in both China and the US, including teams at major tech firms with far greater resources. Translating a compelling research thesis into commercially viable robotic systems remains a significant leap.
There's also the broader challenge facing the entire embodied AI field: real-world environments are messy, unpredictable, and far harder to navigate than digital ones. No startup has yet cracked general-purpose physical intelligence.
The signal: Sinovation Ventures leading this round underscores a broader bet by China's top AI-focused investors on embodied intelligence as the next major platform shift. DeepCybo's "breakout" trajectory — founded only in 2024 and already securing a significant round — mirrors the accelerating pace at which capital is flowing into companies bridging foundation models and physical robotics, a space where the data pipeline, not just the algorithm, may prove to be the decisive competitive moat.
Read more: pandaily.com
