Software Dependency

• Ubuntu 20.04 LTS
• ROS Noetic

详细使用文档 Y1数据采集使用说明

1. Software installation

Choose one of the following two methods

1.1 Conda:

conda create -n imeta_data_collection python=3.8.10
conda activate imeta_data_collection

pip install numpy
pip install rospkg
pip install opencv-python
pip install omegaconf
pip install pynput
pip install dm_env
pip install h5py
pip install matplotlib

1.2 or create conda environment:

conda env create -f conda_environment.yaml
conda activate imeta_data_collection

2. Data collection:

Data Collection KeyBoard usage:

KeyBoard	Description
S(s)	Start collect current episode
E(e)	End and save current episode
R(r)	Restart collect current episode
Q(q)	exit

2.1 collecte data with hdf5 format

source the ros env first, because we use rosmsg in data collection. if you use python sdk:

cd y1_sdk_python/y1_ros
source devel/setup.bash

if you use c++ sdk:

cd y1_sdk
source devel/setup.bash

if you have only one master arm:

python -m scripts.record_data config_file=cfg/one_master.yaml

if you have two master arms:
python -m scripts.record_data config_file=cfg/two_master.yaml

if you have one master and one slave arm:
python -m scripts.record_data config_file=cfg/one_master_slave.yaml

if you have two master and two slave arm:
python -m scripts.record_data config_file=cfg/two_master_slave.yaml

参数配置

以two_master_slave.yaml配置为例解释:

param	value	Description
task_description	"pick and place"	Language prompts for tasks
num_episodes	100	Number of episodes to collect
save_rate	30	Save data rate, Unit: hz

hdf5_cfg

param	value	Description
save_as_h5	True	save data as hdf5 format
dataset_dir	"data/"	save data directory
task_name	"two_arm_teleop"	save directory name

lerobot_dataset_cfg

param	value	Description
save_as_lerobot	True	save data as LeRobotDataset v2.1 format
repo_id	"y1/two_arm_teleop"	repository id
robot_type	"y1"	robot type

robotic_arm_cfg

param	value	Description
collection_type	"two_master_slave"	collection type
master_arm_right_topic	"/master_arm_right/joint_states"	right master arm joint state feedback
puppet_arm_right_topic	"/puppet_arm_right/joint_states"	right puppet arm joint state feedback
master_arm_left_topic	"/master_arm_left/joint_states"	left master arm joint state feedback
puppet_arm_left_topic	"/puppet_arm_left/joint_states"	left puppet arm joint state feedback

camera_cfg

param	value	Description
camera_names	["cam_right_wrist", "cam_front", "came_left_wrist"]	Names of cameras at different location
img_right_topic	"/camera_right/color/image_raw"	right wrist camera rgb image topic
img_left_topic	"/camera_left/color/image_raw"	left wrist camera rgb image topic
img_front_topic	"/camera_front/color/image_raw"	front camera rgb image topic

2.2. collecte data with LeRobotDataset V2.1 format

2.2.1 install lerobot in current python env "imeta_data_collection"

if you want use LeRobotDataset V2.1, install lerobot first:

git clone https://github.com/imeta-lab/imeta_lerobot.git  
cd imeta_lerobot/lerobot  
pip install -e .

2.2.2 set save_as_lerobot to True in yaml file

save_as_lerobot: True

2.2.3 use the program in 2.1 to collect data

python -m scripts.record_data config_file=cfg/your_example.yaml

2.2.4 or convert hdf5 data to lerobot (Recommended)

python -m scripts.convert_h5_to_lerobot --config.raw-dir=data/pick_and_place/ --config.repo-id=y1/pick_and_place

3. Data visualization

3.1 visualize hdf5 data:

python -m scripts.visualize_h5_episode --dataset_dir data/pick_and_place/ --episode_idx 0

3.2 visualize lerobot data:

Refer to the script of lerobot:

'lerobot.scripts.visualize_dataset.py`

you can use h5dump to view hdf5 data:

Show only the file structure (no data):

sudo apt install hdf5-tools

h5dump -H example.h5