Benchmarks & EvaluationReward ModelingCommercial OK
rbm-1m-ood-full
by aliangdw
1.4Kdownloads
0likes
Description
RBM-1M-OOD evaluation dataset used in Robometer. It contains over 1k trajectories used for evaluation of general-purpose reward models.
Dataset Description
Official evaluation in the paper uses only these 6 data sources: usc_trossen, mit_franka, utd_so101, usc_xarm, usc_franka, usc_koch. Reported benchmarks and metrics in the paper are computed on this subset.
The repository may also include trajectories from additional data sources (e.g. utd_so101_wrist, usc_koch_paired… See the full description on the dataset page: https://huggingface.co/datasets/aliangdw/rbm-1m-ood-full.
What can I do with this?
Tags
task_categories:roboticslicense:apache-2.0size_categories:1K<n<10Kmodality:textmodality:videolibrary:datasetslibrary:mlcroissantarxiv:2603.02115region:usrobometerrbm-1m-oodreward-modelevaluation