Humanoid robotics advance through AI-driven motion and global data collection

The humanoid robotics industry is seeing significant advancements in motion control and data collection. The Omni humanoid robot has gained viral attention for its unique running style, characterized by a forward-leaning posture with hands held near the face. Engineers clarified that this “shy run” was not a programmed aesthetic but an emergent behavior discovered by AI through reinforcement learning. By minimizing arm movement, the AI reduced the load on shoulder motors and prevented overheating, ultimately allowing the robot to win a 400-meter sprint competition.

Simultaneously, US-based robotics unicorn Figure AI has launched “Index,” a global data collection initiative. Through a mobile app, users can earn rewards by uploading videos of daily tasks—such as folding laundry or cleaning—to train the company’s Helix vision-language-action (VLA) model. Figure AI aims to accelerate the development of embodied AI for domestic and industrial use, with plans to invest over $1 billion in data collection and computing infrastructure over the next year. The company’s goal is to develop robots capable of performing complex household chores through natural language commands.