Open Models Almanac
Methods

specific method · not yet filed

Action chunking

Action chunking predicts and executes a sequence of future actions without intermediate replanning.1 It improves policy success rates across manipulation tasks, but under autoregressive generation the extra decoder passes multiply latency, so it needs parallel decoding to be practical.2,3

Trade-off Chunking raises latency slightly on its own, but combined with parallel decoding it dramatically improves throughput over the baseline.4

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. predicting and executing a sequence of future actions without intermediate replanning — Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
  2. action chunking—i.e., predicting and executing a sequence of future actions without intermediate replanning—improves policy success rates across many manipulation tasks — Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
  3. OpenVLA’s autoregressive generation scheme makes action chunking impractical — Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success
  4. Adding action chunking () increases latency by 17% due to longer attention sequences in the decoder, but when combined with parallel decoding, it dramatically improves throughput — Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success

Also called chunks of K actions.

sources
4
models
3
labs adopt it
3
strongest
core

Documented in

Allen Institute for AI
OpenVLA project (Stanford et al.)

Evidence

5 quotes from 4 sources, one card per model, each opening with its strongest.

How sources treat it
mentioned 1core 4

OpenVLA / OpenVLA-OFT

OpenVLA project (Stanford et al.)adopted2 sources

to predict actions for multiple future timesteps, we simply insert additional empty action embeddings in the decoder’s inputs, which are then mapped to a chunk of future actions

coreunclearin OpenVLA-OFT
2 more from this model

outputs chunks of K actions instead of single-timestep actions

coreunclearin OpenVLA-OFT

incorporating action chunking and temporal smoothing, as implemented in Diffusion Policy, may help OpenVLA attain the same level of dexterity

mentionedother