* feat(logging): add swanlab backend + evo1 stage2 optimizations
- Add SwanLab logging support (auto-selects when SWANLAB_API_KEY is set)
- Add stage2 training optimizations: tensor fastpath, gradient checkpointing
- Document stage2 config override behavior and runnable value-model alternative
Cherry-picked from: bfdc28c, bcecabb
* refactor(logging): separate SwanLabLogger from WandBLogger
- Restore WandBLogger to its original clean form with all comments
- Add SwanLabLogger as a separate class with the same interface
- Add make_logger() factory function to select backend based on
SWANLAB_API_KEY environment variable
- Move evo1 stage2 runtime logging from lerobot_train.py into
EVO1Policy.__init__ where it belongs
- Add ImportError handling for swanlab dependency
* fix(logging): gracefully fall back to WandB when swanlab is not installed
If SWANLAB_API_KEY is set but the swanlab package is not installed,
make_logger() now logs a warning and falls back to WandBLogger instead
of crashing with an ImportError at training startup.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Elvinyk <223015152@link.cuhk.edu.cn>
Co-authored-by: MrBlue0632 <2028207642@qq.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Public training code should be policy-agnostic. This removes:
- resume_pretrain config field (pretrained_path without --resume already
provides the same "load weights, fresh optimizer" behavior)
- evo1-specific DDP static graph and stage2 logging in lerobot_train.py
- drop_last sourced from policy config (DataLoader concern, not policy)
- type-inference fallback in PreTrainedConfig that was added to handle
evo1 checkpoints missing a "type" field
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The recording completion hint should point users to the dataset report
tool first, which itself already suggests boundary frame export.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat: add episode boundary frame export script
* feat: add boundary frame export hints after recording and in dataset report
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat: print scp hint when running in SSH session
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: address review findings — deduplicate utils, fix CHW handling, fix --json output
- Replace duplicated resolve_dataset_root/load_info/HF_LEROBOT_HOME with imports from existing modules
- Replace custom tensor_to_pil_image with torchvision.transforms.functional.to_pil_image
- Fix select_camera_key to use names field for CHW/HWC shape disambiguation
- Move dataset report tip inside text-only branch to preserve --json output
- Move prepare_output_dir after dataset construction to avoid premature deletion
- Fix resolve_repo_id to always derive from resolved dataset_root
- Remove unnecessary scripts/__init__.py
- Fix episode spec display in README.txt for non-contiguous selections
- Use episodes.column_names for column containment check
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: shuyuan <shuyuan@shuyuandeMac-mini.local>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The previous condition gated on enable_collector_policy_id (defaults False),
so pure teleop recordings still missed the three complementary_info fields.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- add send-only follower support and skip unnecessary observation fetches
- add bi-piper process isolation and fresh feedback handling for manual control
- split piper and piperx gravity compensation defaults and expand teleop tests
- return the robot to the stored reset pose for both success and failure outcomes
- clarify reset logging so it matches the generalized behavior
- add a controller test covering success-triggered resets
- detect Git LFS pointer URDF early and raise actionable runtime error
- add regression test for LFS-pointer URDF guard
- normalize PiPER/PiPER-X asset modes from 100755 to 100644
- gate PI0/PI0Fast tests on HF token availability for gated models
- update value infer viz tests to current API ownership and arguments
- align piper calibration assertion with runtime error message
- set HF_HUB_DISABLE_XET in fast_tests workflow
- set piper calibration_mode default to required and keep off as explicit opt-out
- remove auto pairing policy and simplify runtime calibration gating
- update README and add test for uncalibrated control when calibration_mode=off
- document MasterSlaveConfig linkage_config semantics (0xFA/0xFC) in README
- update piper leader startup role command from 0xFA to 0xFC
- add inline note on commandability and power-cycle requirement after role switch
- align piper leader config comment with 0xFC startup role
- add piper_follower robot and piper_leader teleoperator implementations based on Piper SDK CAN APIs
- wire piper types into teleoperate/record/replay/calibrate command imports and device factories
- add Piper SDK import helper with auto-discovery and unit conversion utilities
- add piper teleop tests and camera-compatible follower observation support for record/hil flows