Add a New Experiment Tracker: SwanLab

[YushunXiang]

This pull request introduces support for SwanLab as an experiment tracker alongside WandB in the training pipeline. Key changes include the addition of a SwanLabLogger utility, updates to configuration classes, and modifications to the training script to integrate SwanLab logging.

SwanLab Integration:

• lerobot/common/utils/swanlab_utils.py: Added a new utility module containing the SwanLabLogger class for logging metrics, policies, and videos to SwanLab. This includes helper functions for generating group names, retrieving run IDs, and sanitizing artifact names.
• pyproject.toml: Added swanlab>=0.6.4 as a dependency to enable SwanLab integration.

Configuration Updates:

• lerobot/configs/default.py: Introduced a new SwanLabConfig class for SwanLab-specific configurations, such as enabling/disabling artifacts, specifying project details, and setting the tracking mode. Updated WandBConfig with a deprecation notice for enable and additional fields for better compatibility with the new tracking system.
• lerobot/configs/train.py: Updated TrainPipelineConfig to include a tracker field for selecting between "wandb," "swanlab," "both," or "none." Added validation logic to ensure valid tracker selection and enable the appropriate configuration dynamically. [1] [2]

Training Script Enhancements:

• lerobot/scripts/train.py: Modified the training script to initialize and use SwanLabLogger based on the selected tracker. Added support for logging metrics, policies, and videos to SwanLab during training and evaluation phases. [1] [2] [3] [4]

Usage Example:

python lerobot/scripts/train.py \
    --dataset.repo_id=lerobot/test  \
    --dataset.root=your-local-path \
    --output_dir=output/test \
    --policy.device=cuda \
    --policy.type=diffusion \
    --policy.push_to_hub=false \
    --steps=1000000 \
    --save_freq=20000 \
    --num_workers=8 \
    --batch_size=16 \
    --tracker=swanlab \
    --swanlab.project=your-project \
    --swanlab.run_id=test \

Set tracker=swanlab

---

If you have a plan to merge this PR, I will update the document in detial. @aliberts

[Copilot]

Pull Request Overview

This pull request introduces SwanLab integration as a new experiment tracker option alongside WandB in the training workflow. It adds a new SwanLabLogger utility, updates configuration classes to support the new tracker, and modifies the training script to initialize and use SwanLab logging when selected.

• Added SwanLabLogger in lerobot/common/utils/swanlab_utils.py to handle logging of metrics, policies, and videos to SwanLab.
• Updated pyproject.toml with the swanlab dependency and modified configuration files to support dynamic tracker selection.
• Enhanced lerobot/scripts/train.py to initialize and utilize SwanLabLogger based on the chosen tracker.

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
pyproject.toml	Added swanlab dependency
lerobot/scripts/train.py	Updated logger initialization and logging calls to include SwanLab
lerobot/configs/train.py	Updated config with tracker field and validation logic
lerobot/configs/default.py	Added SwanLabConfig and updated WandBConfig with deprecation notices
lerobot/common/utils/swanlab_utils.py	Introduced SwanLabLogger and helper functions for SwanLab integrations

[Copilot]

Consider using match.group(1) instead of match.groups(0)[0] for clarity when extracting the run ID.

Suggested change

	swanlab_run_id = match.groups(0)[0]
	swanlab_run_id = match.group(1)