Collected sources and patterns will appear here. Add from search or the patterns library.
RawActionTrajectory -> NormalizedActionTrajectory
Map raw robotic action limits (e.g. cartesian or joint coordinates) to a normalized scale (such as [-1, 1]) prior to diffusion head inference, and invert the mapping on generation.
Problem it solves
Direct regression of raw continuous control parameters across heterogeneous robot tasks introduces training instability.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.