Open Models Almanac
Methodsinference & servingdecoding strategy

single mention · inference & serving

Second-most-likely action querying

Second-most-likely action querying is a decoding workaround for robot policies that returns the model's second-ranked action instead of its first.1 It is used when the top action is degenerate — often all zeros — causing the robot to freeze.2

Trade-off It improves rollout performance but is suboptimal compared to retraining on preprocessed data.3

How this definition was written

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

  1. we mitigated this issue by simply querying the second-most-likely action from the model — OpenVLA: An Open-Source Vision-Language-Action Model
  2. the first-most-likely action was often all zeros while the second-most-likely action was not — OpenVLA: An Open-Source Vision-Language-Action Model
  3. though we believe that it is still suboptimal compared to re-training the model on the preprocessed version of the dataset — OpenVLA: An Open-Source Vision-Language-Action Model

Also called Always query the second-most-likely action, querying the second-most-likely action.

Filed under inference & serving › decoding strategy