由于采集玩数据都要上传到hugging face,除了墙的原因还有就是笔记本的摄像头会拍到我的帅脸所以不想上传.本来一开始想说先最快跑通测试.但是还是花时间研究了一下代码通通改成本地处理. 指令如下 .. 代码修改放到下一篇.
对了 斜杠号不知为啥被消除了,参照图片加进去吧
[采集数据] pda1可以换成你要训练的名称,我因为抓的是熊猫布偶所以取这个名称.
python lerobot/scripts/control_robot.py record
--robot-path lerobot/configs/robot/so100.yaml
--root ./outputs/pda1/
--fps 30
--warmup-time-s 5
--episode-time-s 40
--reset-time-s 10
--num-episodes 20
--push-to-hub 0
--local-files-only 1
--single-task real
[数据可视化]
python lerobot/scripts/visualize_dataset.py
--repo-id lerobot/so100
--root ./outputs/pda1/
--local-files-only 1
--episode-index 1
[训练]
export HYDRA_FULL_ERROR=1
python lerobot/scripts/train.py
policy=act_so100_real
env=so100_real
hydra.run.dir=train/pnd1/
hydra.job.name=act_so100_bear
device=cuda
wandb.enable=false
[复现]
python lerobot/scripts/control_robot.py record
--robot-path lerobot/configs/robot/so100.yaml
--tags so100 tutorial eval
--fps 30
--warmup-time-s 5
--episode-time-s 30
--reset-time-s 30
--num-episodes 3
--single-task real
-p train/bear01/checkpoints/last/pretrained_model
