Reasoning models still need help to control humanoids, one post argues

One post describes Gemini Robotics 2 as placing an embodied reasoning model above a vision-language-action model for whole-body control, and Anthropic as testing general-purpose reasoning models at several levels of control. The author argues that general-purpose reasoning models cannot yet control a humanoid on their own and need a pretrained control policy underneath. Still, they think a model with strong multimodal and reasoning capabilities could eventually do so, describing HumanCLAW as “our attempt to get an early glimpse of that future.”