MultiLeRobotDataset' object has no attribute 'meta

[astroyat]

Encounter the following issue when training with multiple dataset repo, modify the code to remove meta to workaround but then not compatible with single dataset.

DATA_DIR=data python lerobot/scripts/train.py
dataset_repo_id=[astroyat/cube,astroyat/eval_cube]
policy=act_csv_real
env=so100_real
hydra.run.dir=outputs/train/act_eval_cube
hydra.job.name=act_eval_cube
device=cuda

Error executing job with overrides: ['dataset_repo_id=[astroyat/cube,astroyat/eval_cube]', 'policy=act_csv_real', 'env=so100_real', 'device=cuda']
Traceback (most recent call last):
File "/home/target/lerobot/lerobot/scripts/train.py", line 652, in train_cli
train(
File "/home/target/lerobot/lerobot/scripts/train.py", line 313, in train
offline_dataset = make_dataset(cfg)
File "/home/target/lerobot/lerobot/common/datasets/factory.py", line 114, in make_dataset
dataset.meta.stats[key][stats_type] = torch.tensor(stats, dtype=torch.float32)
AttributeError: 'MultiLeRobotDataset' object has no attribute 'meta'

Traceback (most recent call last):
File "/home/target/lerobot/lerobot/scripts/train.py", line 652, in train_cli
train(
File "/home/target/lerobot/lerobot/scripts/train.py", line 331, in train
dataset_stats=offline_dataset.meta.stats if not cfg.resume else None,
AttributeError: 'MultiLeRobotDataset' object has no attribute 'meta'

[andlyu]

Had a similar error when trying to run training with 2 datasets. Any luck with this?

Small update: Just replaced dataset.meta.stats with dataset.stats where it was giving errors. Messy but worked for the time being.

[astroyat]

Had a similar error when trying to run training with 2 datasets. Any luck with this?

Small update: Just replaced dataset.meta.stats with dataset.stats where it was giving errors. Messy but worked for the time being.

I use this

if isinstance(dataset, MultiLeRobotDataset):
dataset.stats[key][stats_type] = torch.tensor(stats, dtype=torch.float32)
else:
dataset.meta.stats[key][stats_type] = torch.tensor(stats, dtype=torch.float32)