Tedrake: Long-Context Policies May Outperform Action Chunking

RussTedrake · x · 2026-08-21

Russ Tedrake reflected on Diffusion Policy, citing a new paper that suggests the effectiveness of action chunking is partly due to the short observation context lengths typically used. The paper argues that long open-loop execution helps short-context policies imitate non-Markovian experts. Tedrake finds this compelling evidence for shifting to policies with long context lengths and short action horizons.

Original post →

More from Embodied

Embodied channel →