AI 导读
中国的机器人格斗场景又升级了。一段新演示显示，Unitree 的 G1 能连贯衔接动作，并以惊人的速度做出反应。 它的世界模型读取视觉输入，预测接下来会发生什么，并从预定义动作库中选出最佳响应。 G1 仍无法自主规划新动作。但说实话，这正是机器人格斗开始让人兴奋的地方：还不完全智能，但已经足够快、反应灵敏，越来越难以被忽视。
正文
China’s robot-fighting scene just leveled up. A new demo shows Unitree’s G1 chaining moves together and reacting at remarkable speed.
Its world model reads visual input, predicts what comes next and chooses the best response from a library of predefined actions.
The G1 still cannot plan new movements autonomously. But honestly, this is where robot combat starts to feel exciting: not fully intelligent yet, but already fast, responsive and getting harder to dismiss.
来源：@thexpin · x.com