search

BarisYazici / deep-rl-grasping

Train deep reinforcement learning model for robotics grasping. Choose from different perception layers raw Depth, RGBD and autoencoder. Test the learned models in different scenes and object datasets
MIT License
203 stars 46 forks source link

The results of your experiment are so wonderful #23

Closed YKJK closed 1 year ago

YKJK commented 1 year ago

I have some Deep rl grasping models question, How to modify the program code? 1.Can the camera be changed to a fixed position? 2.I would like to know how to adapt this model for use in a real environment. 3.Is it possible to change the gripper to match the shape of the KUKA youBot gripper? 4.Could you provide instructions on changing the model of the grasped object?

BarisYazici commented 1 year ago

Sorry for replying with delay.

Camera position can be chaged. Try changing this https://github.com/BarisYazici/deep-rl-grasping/blob/fd2214e42a77e80fa986f04e27e043e0133b4d51/config/camera_transform.yaml#L1C1-L1C1

  1. I haven't try in the real environment. But I guess you need to apply domain randomization during training. You can implement an interface for that sucht that the Pybullet simulation physics parameters are chaged. Contributions are welcomed :)

  2. You can easily change the gripper or the robot look at other branches there should be examples of using the Franka and Kuka robot. In general you need to have a robot description file. https://github.com/BarisYazici/deep-rl-grasping/blob/fd2214e42a77e80fa986f04e27e043e0133b4d51/config/full_depth_obs.yaml#L3

  3. There is an example of using different object models in the repo -> https://github.com/BarisYazici/deep-rl-grasping/tree/fd2214e42a77e80fa986f04e27e043e0133b4d51/models/wooden_blocks. You need to have a similar structure. And change the config.yaml file to upload the models https://github.com/BarisYazici/deep-rl-grasping/blob/fd2214e42a77e80fa986f04e27e043e0133b4d51/config/full_depth_obs.yaml#L15C1-L15C25

YKJK commented 1 year ago

Thank you for your instant response. I will ask you again if I have any related questions!!