Hi @suraj-nair-1 , @niccolofusai13 , @jimmyt857,
Thanks for releasing the ros code modifications for running openpi on ALOHA. I am using a stationary ALOHA-1 platform. I had my data collected using the original 4arms_teleop.launch script. I realized that the inference code for openpi, under openpi/third_party/aloha/, particularly files realsense_publisher.py and ros_nodes.launch changes (i) camera FPS from 60 to 30 Hz and (ii) changes the default image processing settings such as auto-exposure, focus and raw pixel format. As a result, my raw camera data being fed to the policy during training and inference has very different image processing and control frequency. I wonder if pi works better on ALOHA with these changes or other reasons for the changes, and what is recommended for good results.