Unitree Develops Autonomous Feedback Loop For Humanoid Robots
Unitree Robotics is developing a self-evolving feedback loop designed to allow humanoid robots to write code, run simulations, and improve their performance independently. Founder Wang Xingxing unveiled the project, dubbed Unitree Rehearsal.
The system shifts development away from labour-intensive manual programming towards a continuous and autonomous feedback loop. Artificial intelligence will search research papers, write code, test it in simulations, and run trials on physical robots.
Wang noted that the ultimate goal is creating an automated cycle that can exponentially accelerate the development of robotics algorithms. Robots must navigate environmental unpredictability and mechanical errors unlike digital software.
Videos shown at the presentation demonstrated Unitree robots using OpenAI's GPT-6 model to map routes and plan movements without task-specific training. Humanoid robots could potentially tidy rooms on command as early as 2027 or 2028.
Tesla Chief Executive Officer Elon Musk recently stated that humanoid robots could eventually become the biggest product on Earth and outnumber humans. Tesla still faces about a year of development hurdles with its Optimus robot.
Data indicates a challenging transition from laboratory to living room environments. Stanford University evaluations found that robots succeeded in only 12% of real-world household tasks despite performing significantly better in simulation.
* Unitree Robotics is creating a self-evolving feedback loop called Unitree Rehearsal.
* The project aims to replace labour-intensive manual programming with artificial intelligence that writes code, runs simulations, and tests physical robots.
* Humanoid robots could potentially begin tidying rooms on command by 2027 or 2028.
* Stanford University evaluations revealed that robots succeeded in only 12% of real-world household tasks.
Source: SCMP
