One Brain, Multiple Robots

Using a single embodied-AI model to control many different kinds of robot hardware.

“One brain, multiple robots” is a phrase used in China's robotics industry for training a single general-purpose robot “brain” model that can drive robot arms, dual-arm rigs, wheeled platforms, humanoids, and quadrupeds alike, instead of training a separate model for each type of machine. Technically, it relies on cross-embodiment training: data from many different robots is pooled for pretraining, and a unified action space or embodiment-specific action heads then handle differences in joint count and control scheme. The appeal is that data can be reused across robots and a model company isn't locked into one hardware platform; the difficulty is that embodiments differ enough that negative transfer — where training on one robot hurts performance on another — is common. Both “brain” companies and the model–hardware decoupling business model rest on this premise.

ExampleSkild AI says its Skild Brain can control multiple robot embodiments; Physical Intelligence's π0 is jointly trained on data from several different robots.

Also called

One Brain, Multiple Embodiments, One Brain, Multiple Bodies

Related

Cross-Embodiment · Embodiment Gap · Unified Action Space · Embodiment-Specific Head · Robot-Brain (Model-Only) Company · Hardware-Software Decoupling

Sources

Skild AI
π0: Our First Generalist Policy - Physical Intelligence

As of

2025