Why Giving Up the Remote Is Embodied Intelligence’s Final Hurdle

Summary

Robot competitions are showing a clear shift from remote control toward autonomous operation. In the second robot half-marathon, the autonomous robot “Lightning” finished in 50 minutes 26 seconds, nearly 1 hour 50 minutes faster than the previous champion and ahead of the human half-marathon record cited in the article; 40% of entrants ran autonomously. The scoring system gave remote-controlled entries a 1.2 time multiplier, while 23 of 30 events at the second World Humanoid Robot Games required full autonomy. The article explains that a remote controller usually issues task-level commands, while the robot handles locomotion, balance, obstacle avoidance, and other low-level actions itself, so this is different from operating a toy car. Remote control remains a practical transition mechanism because it provides safety and produces real-world training data; the games reportedly accumulated 2,500 hours across 12 scenarios, 44 tasks, and more than 100 skills. The main technical barrier is generalization: robots can approach a 100% success rate in familiar settings but perform much worse in unfamiliar environments. Wang Xingxing’s proposed “double 80%” standard requires a robot to complete about 80% of tasks in 80% of unfamiliar scenarios from voice or text instructions. The article places current humanoid robots around the early L2 stage and describes the path toward L3 and beyond as closing the gap between high-level intent and coordinated physical action. It also cites autonomous training behind Tiangong Omni’s distinctive gait and expects future competitions to make full autonomy a hard scoring requirement. Wang estimates that a “ChatGPT moment” for robots could arrive in two to three years in an optimistic case, or five to ten years otherwise, but the article presents this as a forecast rather than an established result.