Robots Learn to Code Themselves: Self-Evolving AI Arrives

The field of robotics is experiencing a period of significant innovation, with humanoid robots rapidly improving their capabilities in task execution, navigation, and human interaction. Key players like Unitree, Tesla, and Figure have recently shared substantial updates on their progress. Unitree is pioneering an AI system called 'Unitree Rehearsal,' designed to enable its humanoid robots to autonomously improve their own code and simulations, reducing the need for manual human intervention and accelerating algorithm development. Wang Xingxing, Unitree's founder, anticipates these robots could be cleaning rooms as early as next year.

Tesla has commenced production of its Optimus robots at a repurposed electric vehicle factory in Fremont, California. Elon Musk projects a potential ramp-up to 20,000 robots per week by 2027, contingent on further engineering. While production lines are operational, engineers continue to refine the robot's design for enhanced safety and reliability, addressing current limitations in independent task completion that require careful environmental assessment and movement planning.

Figure, another major robotics company, has reported significant advancements in its fourth-generation humanoid robots. The company is scaling production of its Figure 03 model and has achieved millimeter-level precision in fingertip tracking, bringing its robots closer to human dexterity. Figure is also leveraging extensive data from human activities, with over 37.5 million videos uploaded to its Index app for robot training, alongside durability upgrades like a Figure 03 bot operating continuously for nine days on a treadmill.