Introducing Streaming Diffusion Policy (SDP)! SDP allows you to generate actions using a single denoising step, with no distillation! SDP keeps an action buffer of partially denoised future actions, which is refined over time. Led by Sigmund Hennum Høeg

@sigmundhh it's so beautiful

@sigmundhh Very neat! The connection to "uncertainty" also makes a lot of sense in this setting. Actions "far into the future" are harder to predict, as the state you'll be in is less clear (e.g., stochastic dynamics). So you only start "committing" to an action as it becomes close.

@sigmundhh Did you guys try something mctg-like from Diffusion Forcing with this method to see if there's a performance benefit?

@sigmundhh We didn’t get a chance — that would be cool to do!

@sigmundhh Both this and Diffusion Forcing have literally opened doors for me to revisit an old project and possibly complete it. Great work @du_yilun