Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
Frame -> StackedState
Combine the current observation frame with the preceding N-1 frames into a single multi-channel tensor, zero-padding if the episode has fewer than N frames.
Problem it solves
Single visual frames lack the temporal velocity and direction context needed to satisfy the Markov property.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.