EDI Ros Pipeline

Notice

In the following part, if you are executing on (pc@lab), you need not to run source /home/pc/ediControler/catkin_ws/devel/setup.bash

TODO

In the gym obs, keys are "status" and "images' (deprecated soon).

In the lmdb dataset, keys are "status" and "sensors'.

Install (New)

Compile ros rpc server dependency.

cd catkin_ws
catkin_make -DPYTHON_EXECUTABLE=/usr/bin/python3

How to run

Setups

This section includes ROS Configurations and starting several hardware interfaces.

1. Set hosts with on both master (pc@lab) and slaves

Add to your hosts file with sudo vim /etc/hosts:

192.168.1.240   lab
192.168.1.250   timetserver
192.168.1.38    radiance_wired
192.168.1.147   radiance_wireless

2. Run start_publishing.sh on master (pc@lab).

    It will firstly `source /home/pc/ediControler/catkin_ws/devel/setup.bash` and then start nodes below:
    ```
    Camera Nodes
    Status Node
    Real Environment Control Backend
    Sim Environment
    ```
   

3. Set environment variable on your computer.

export ROS_HOSTNAME=timetserver # Replace this hostname
export ROS_MASTER_URI=http://192.168.1.240:11311 # Do not Replace this one

Policy Inference

4. Add edi_gym directory to your directory, Or you can add a link to the site-package path.

echo /home/radiance/projects/ediControler/  > /home/radiance/miniconda3/envs/ml/lib/python3.8/site-packages/edi_gym.pth

5. (New) Run source /home/pc/ediControler/catkin_ws/devel/setup.bash

6. Run a test example python edi_env_example.py.

Human Demonstration

4. (New) Run source /home/pc/ediControler/catkin_ws/devel/setup.bash

5. Run demo arm python hardware/arm_demo.py.

How to record

1. (New) Run source /home/pc/ediControler/catkin_ws/devel/setup.bash

2. Run recording backend python data_collection/recorder.py.

   • This command initiates a background data collection program. In the script, you can modify two parameters: lmdb_save_path_is_fixed and delete_bag.
     • lmdb_save_path_is_fixed: If set to True, all episodes from this session will be stored in a single database directory. If set to False, a new database directory will be created for each episode.
     • delete_bag: This parameter determines whether the recorded rosbag files are deleted.

3. Run recording frontend (keyboard input) python data_collection/keyboard_record_trigger.py.

Recording backend will be integrated to start_publishing.sh in the future.

File Organization

The dataset is organized into the following directory structure for efficient access and management:

• dataset/
  • bag/
    • This directory contains the bag files. These are typically used for storing and transporting a collection of files.
  • train_xxx_lmdb/
    • This directory contains the LMDB (Lightning Memory-Mapped Database) files for training. LMDB is used for its high performance and efficiency in reading/writing large data sets.

How to replay

Please check the python files for passing arguments.

Replaying rosbag file (experimental)

1. (New) Run source /home/pc/ediControler/catkin_ws/devel/setup.bash

2. Run python data_collection/bag_loader.py -i -a .

Replaying lmdb file (experimental)

1. (New) Run source /home/pc/ediControler/catkin_ws/devel/setup.bash

2. Run python data_collection/lmdb_interface.py -i -a .

Run with Docker on timetserver

To run in a docker container, you need to run

CUDA="--gpus all -e NVIDIA_DRIVER_CAPABILITIES=compute,utility -e NVIDIA_VISIBLE_DEVICES=all "
sudo docker run -it --rm --net=host --shm-size 32G \
-v <your directory>:/workspace/<docker directory> \
$CUDA edi/ros