The dataset viewer is not available for this split.
Error code: FeaturesError
Exception: StopIteration
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 244, in compute_first_rows_from_streaming_response
iterable_dataset = iterable_dataset._resolve_features()
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 4645, in _resolve_features
features = _infer_features_from_batch(self.with_format(None)._head())
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2866, in _head
return next(iter(self.iter(batch_size=n)))
StopIterationNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Estimate, don't imitate: paper checkpoints
Trained students and frozen experts behind Tables II–IV of Estimate, don't imitate
(ICRA 2026). Code: HHI-Robot-Learning/26-DistillingSGRL
(tag paper-icra26); simulator data: Shcherba/csrl-robohub-data.
One training seed (42) per reported cell. The paper's ± values average seeds 42, 1, 2.
manifest.csv: one row per table cell →run_dir, md5 of the checkpoint, and the eval results recorded for it (1000 episodes, eval seed 420).runs/<family>/<run>/:latest_checkpoint.pth(the final checkpoint),.hydra/overrides.yaml(the exact training recipe), training logs, eval CSVs. A run that backs several cells is stored once.experts/: the five frozen TD7 experts (*.torch), their resolved configs, the start/goal pools (configs/), and the exported actors the estimator loads (deployed_exports/, including the one deployed on the real robot).
Evaluate a checkpoint with the code repo:
uv run python scripts/evaluate.py runs/tagcube_battery/<run> --episodes 1000 --seed 420 --scenario <scenario> --set ...
Caveat (pixel imitation: flow, ACT, MIP). The published numbers for these columns were
evaluated on the best-validation checkpoint, which was not archived. The checkpoints here
are the final ones, so re-evaluating them will not reproduce those cells exactly. The
estimator and full-state columns were evaluated on the final checkpoints provided here.
Eval CSVs under evals/latest/ name best_checkpoint.pth only because the final
checkpoint was evaluated under that file name.
- Downloads last month
- 155