Cao Xi and Zhiyuan Invested in a Post-95s Founder
Current Robotics, a humanoid intelligence company, recently disclosed it has completed seed, angel, and pre-A funding rounds, raising a total of several hundred million RMB. Investors include prominent institutions like BV Baidu Ventures, Hillhouse Capital, Oasis Capital, Monolith, and Qianhai Ark, as well as strategic partners like Zhijin (Agibot), Xinghai Map, and Jike Technology.
The founder behind the company is Zhu Yichen, a talented individual born in the 1990s and a former head of embodied AI at Midea Group. He and his team are pioneers in China for early research on VLA and world models. Their work, cited by Physical Intelligence, laid the technical foundation for Current Robotics.
Current Robotics focuses on a key gap in embodied AI: **Whole-Body Dexterous Manipulation**. Traditional approaches often separate mobility and manipulation, but real-world tasks require seamless, coordinated movement. Their solution, the base model **Curr-0**, integrates navigation, balance, and dexterous hand control into a single, end-to-end trained policy, enabling robots to perform tasks like moving objects through doorways or clearing a table while in motion.
To address data scarcity, the company developed a self-researched wearable data collection system (**HumanEx**) that captures human motions, forces, and visual perspectives in real-world settings like homes and offices, providing rich, scalable, and cost-effective physical data.
For rapid evaluation and iteration, Current Robotics has pioneered the use of world models for strategy assessment and post-training. Their recently released **CurrentWorld-0** is an interactive world simulator that supports multi-robot platforms, multi-camera perspectives, and force/tactile prediction. It allows for policy testing in simulated environments, identification of failure modes, human-in-the-loop correction, and subsequent training with the generated corrective data, forming a complete feedback loop.
Current Robotics aims to build an infrastructure for embodied intelligence, creating a closed-loop system that connects real human behavior data, whole-body robotic skill learning, and efficient world model-based evaluation and improvement. This approach is designed to accelerate the path for robots to move from demonstrations to performing stable, useful work in the complex, real world.
marsbit08/20 02:56
