📰 DeepSeek is reportedly planning to lead a seed round for robotics AI startup Winverse AI, raising tens of millions of dollars and valuing the company at nearly $400 million. What’s more unusual is that Winverse was founded only a few months ago and is still operating in stealth mode, with little information disclosed about its product, team size, or specific business.
🔥 Founder Hu Han has an impressive résumé: he has worked at Baidu’s Institute of Deep Learning and Microsoft Research Asia, and led teams at Tencent working on language, vision-language, and embodied models. He also contributed to the Swin Transformer, which won the 2021 ICCV Marr Prize. Reports differ, however, on whether Winverse is focused on robotics AI or on underlying infrastructure such as model acceleration.
💡 Seen alongside DeepSeek’s recent moves, this investment is about more than backing a new company. In August this year, DeepSeek participated in Unitree Robotics’ IPO strategic placement, investing approximately 140.76 million yuan and receiving 933,399 shares. The two sides also agreed to collaborate on model training, robotics, and embodied intelligence research and development. One is focused on robotics hardware, while the other has a background in model research—their plans appear to be expanding beyond developing models alone and toward real-world devices and applications.
👀 But a large model’s ability to understand images doesn’t mean it can already make a robot act reliably. Spatial reasoning, path planning, motion control, and real-time adaptation to changing environments remain separate challenges. DeepSeek has previously released the vision-language model DeepSeek-VL2, and its robotics collaboration may offer hardware and real-world interaction environments. But there’s still a long way to go before control problems are truly solved.
🤔 DeepSeek is now engaging with both robotics companies and embodied intelligence teams, sending a fairly clear signal. But Winverse hasn’t announced a product yet, and it will take time to see whether this investment leads to tangible technical results. Which part of this strategy do you think is more likely to deliver results first: partnering on robotics hardware or backing a model team?
#DeepSeek #具身智能 #机器人AI #ArtificialIntelligence
🔥 Founder Hu Han has an impressive résumé: he has worked at Baidu’s Institute of Deep Learning and Microsoft Research Asia, and led teams at Tencent working on language, vision-language, and embodied models. He also contributed to the Swin Transformer, which won the 2021 ICCV Marr Prize. Reports differ, however, on whether Winverse is focused on robotics AI or on underlying infrastructure such as model acceleration.
💡 Seen alongside DeepSeek’s recent moves, this investment is about more than backing a new company. In August this year, DeepSeek participated in Unitree Robotics’ IPO strategic placement, investing approximately 140.76 million yuan and receiving 933,399 shares. The two sides also agreed to collaborate on model training, robotics, and embodied intelligence research and development. One is focused on robotics hardware, while the other has a background in model research—their plans appear to be expanding beyond developing models alone and toward real-world devices and applications.
👀 But a large model’s ability to understand images doesn’t mean it can already make a robot act reliably. Spatial reasoning, path planning, motion control, and real-time adaptation to changing environments remain separate challenges. DeepSeek has previously released the vision-language model DeepSeek-VL2, and its robotics collaboration may offer hardware and real-world interaction environments. But there’s still a long way to go before control problems are truly solved.
🤔 DeepSeek is now engaging with both robotics companies and embodied intelligence teams, sending a fairly clear signal. But Winverse hasn’t announced a product yet, and it will take time to see whether this investment leads to tangible technical results. Which part of this strategy do you think is more likely to deliver results first: partnering on robotics hardware or backing a model team?
#DeepSeek #具身智能 #机器人AI #ArtificialIntelligence