Flexion Robotics 是一家由前 Nvidia 机器人研究人员创立的瑞士新创公司，该公司开发了一种新方法，透过在模拟环境中训练机器人的单一技能，并使用主人工智慧（AI）演算法来决定如何调度这些技能，使人形机器人能够在现实世界中执行如开门、爬楼梯和搬运箱子等复杂的日常工作。

该系统的核心技术在于结合了多种 AI 模型，主模型透过消化人类活动的影片来学习何时采取何种行动，再触发在模拟中学到的具体技能，并在软体的各个层面广泛应用了强化学习，从高层决策、模拟训练到马达控制，皆透过试错来优化机器人的肢体动作与平衡。

行业分析师指出，人形机器人的革命性关键不在于硬体本身，而在于其背后的 AI 模型。Flexion 的软体具备跨多种人形硬体运作的特性，这使其与多家机器人公司的合作具有更高的商业价值，但在市场化过程中仍需与硬体制造商紧密合作以应对激烈的竞争。

Flexion Robotics, a Swiss startup founded by former Nvidia robotics researchers, has developed a new method to train humanoid robots to perform complex chores by teaching them individual skills in simulation and using a master AI algorithm to orchestrate these actions in the real world.

The core technology combines different AI systems where the main model learns what actions to take by digesting videos of human activities, triggers simulated skills, and extensively utilizes reinforcement learning across all software layers to control the robot's motors and maintain balance.

Industry analysts emphasize that the revolutionary value of humanoids lies in their underlying AI models rather than the hardware, and Flexion's hardware-agnostic software increases its commercial potential, though success will require close collaboration with hardware manufacturers in a competitive market.