This Robometer model is full-finetuned on the libero_all dataset containing: LIBERO-90, 10, Object, Goal, Spatial, and generated failures from each of these datasets. This is likely to outperform the default Robometer model because the standard one is not trained on LIBERO-90.

Model Details

Base Model: Qwen/Qwen3-VL-4B-Instruct
Model Type: qwen2_5_vl

Training Run

Wandb Run: rbm
Wandb ID: wj739wca
Project: robometer
Notes: training Robometer

Citation

If you use this model, please cite: