diffuse_to_clone

Diffusion Models for Imitation Learning

Pre-requisites:

1. Install Docker
2. Install Nvidia Docker
3. Work with a system that has the franka arm configured with the frankapy package

Setup:

1. Make Docker Container (use sudo if docker is not added to the user group):

   docker build -t diffusion .
   

   Use newgrp docker if you didnt logout

Running the code:

0. Unlock robot joints

   • ssh -X student@[control-pc-name]
   • google-chrome
   • In firefox open https://172.16.0.2/desk/ and press Click to unlock joints

1. Run the start_control_pc script from the frankapy package

   • cd <frankapy package directory>
   • bash ./bash_scripts/start_control_pc.sh -u student -i [control-pc-name]

2. Run the MoveIt Server

   • Start the docker container

     bash run_docker.sh
     

   • Inside the container:

     roslaunch manipulation demo_moveit.launch 
     

3. Collect data with vision:

   roslaunch manipulation data_collect_vision.launch
   

   Launch realsense from inside PC

   export ROS_IP=172.26.230.217 && export ROS_MASTER_URI=http://172.26.165.201:11311/ && roslaunch realsense2_camera rs_camera.launch 
   

4. SCP command:

   • Transfer Dataset

     • All with scp: scp -r /home/vib2810/diffuse_to_clone/dataset/data/vision_new_camera guest@punisher.wifi.local.cmu.edu:/home/guest/vibhakar/diffuse_to_clone/dataset
     • Transfer only new files (requires to run bash claim_files.sh on punisher) rsync -avz --ignore-existing /home/vib2810/diffuse_to_clone/dataset/data/vision_new_camera guest@punisher.wifi.local.cmu.edu:/home/guest/vibhakar/diffuse_to_clone/dataset/data/
     • Auxilio Brain rsync -avz --ignore-existing /home/vib2810/diffuse_to_clone/dataset/data/vision_new_camera abhinav@172.26.73.171:/home/abhinav/Desktop/diffuse_to_clone/dataset/data/

   • Transfer Models from Punisher to Local - scp -r guest@punisher.wifi.local.cmu.edu:/home/guest/vibhakar/diffuse_to_clone/logs/models/* /home/vib2810/diffuse_to_clone/logs/models/