search

droid-dataset / droid

Distributed Robot Interaction Dataset.
https://droid-dataset.github.io/droid/
108 stars 18 forks source link

Question about cartesian_position in observation and action_dict #2

Open yangsizhe opened 5 months ago

yangsizhe commented 5 months ago

I appreciate the work you've done on this project. I have a question regarding the cartesian_position in the observation and action_dict. Which pose does this cartesian_position refer to? Is it the pose of the TCP (Tool Center Point), which is the position of the light green cube in the image, or is it the pose of the gripper frame (where the red line represents the X-axis and the green line represents the Y-axis)?

Additionally, I'm curious about how the base frame of the robot is defined. When I import joint state data into the simulator, I find that the position of the end effector differs from the position in the cartesian_position in the dataset by approximately [-0.36, -0.44, 0.10].

Thank you in advance! gripper

AlexanderKhazatsky commented 5 months ago

Hi, we experimented with many different grippers and to keep things consistent we use the wrist site of the robot rather that the actual gripper location. I believe that should explain the issue you're seeing. Let me know if you have any follow up questions!

yangsizhe commented 5 months ago

Thank you for your explanation! I would like to know how to conveniently obtain the TCP pose through joint_position, gripper_position, and the gripper model. Do you have any recommended tools, such as corresponding models for hardware in simulator, to solve the TCP pose through forward kinematics?

Nimolty commented 4 months ago

Dear authors,

I am writing to you to ask some questions about the action space in the paper "Droid".

I noticed that in the Github Issues, the author reminded that the action is the "wrist pose". However, I could hardly understand what does the wrist mean? I try to visualize the action pose but couldn't find the "wrist space origin and x,y,z-axis". I am trying to transform the original wrist action pose into the tcp action pose, so the accurate definition of wrist space is quite important.

I have attached a figure and could you please show the concrete definition of wrist space. I would appreciate it very much if you could solve my concerns.

Nimolty commented 4 months ago

image.pdf

AlexanderKhazatsky commented 4 months ago

The wrist site would correspond to the part of the gripper where the red and green lines are protruding. In other words, where it attached to the franka arm.

We've done some internal work a while back to work form the TCP site instead of the wrist site, and I've asked the person who led this effort to get back to you!

AlexanderKhazatsky commented 4 months ago

To answer your question Nimolty, wrist site would be the spot between Link 7 and gripper mount in your diagram