Open Models Almanac
Methodstraining objectiveauxiliary loss

single mention · training objective

Auxiliary object-detection loss

An auxiliary object-detection loss trains a model, alongside its main action prediction, to localize the object of interest from a language instruction.1 A linear layer on the vision-language embedding tokens predicts normalized 2D bounding-box center coordinates, optimized with a squared loss and added to the final loss.2

How this definition was written

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

  1. To enhance the model’s spatial understanding, we introduce an auxiliary object detection loss during training. In addition to predicting actions, the model must also localize the object of interest based on the given language instruction — GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
  2. we append a linear layer atop the final vision-language embedding tokens and optimize using a squared loss — GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

Filed under training objective › auxiliary loss