Gazebo bug with stretch

[NateWright]

When nothing is happening in the simulation stretch will drift off and spin a little in simulation. This is noticeable in a short period of time and even more so when gazebo is left open for longer when you are testing code periodically.
I am using dev/noetic branch.
Time 0 min:

Time 5 min:

Eventually the robot will be facing up against the wall and driving into it.

[hello-binit]

Hi @NateWright, thanks for reporting this. I wonder if the drifting is similar to this issue about the Kobuki diff drive robot: https://answers.gazebosim.org//question/25794/idle-robot-drifts-due-to-tilted-inertia/

The URDF of Stretch loaded into Gazebo has a fixed joint for the caster wheel. It's possible that making it a continuous joint with friction would help with this issue.

[NateWright]

@hello-binit I looked into it a bit and noticed that the caster wheel is built into the base_link.STL file and then there is another wheel attached to the robot in the same position. I am attaching a photo showing the gray wheel from the base_link file and a blue omni wheel colliding.
https://github.com/hello-robot/stretch_ros/blob/dev/noetic/stretch_description/meshes/base_link.STL

I wonder if removing the wheel from the stl file might fix this issue

[hello-binit]

Definitely worth trying. I won't be near a desktop until next week, but you can try importing the mesh into Blender to erase the "fixed" caster wheel.

[cyb1212]

@NateWright I am facing the same problem. Do you solve it? Thank you.

[NateWright]

@NateWright I am facing the same problem. Do you solve it? Thank you.

I messed with it for awhile, but I was unable to solve the issue

[cyb1212]

Thanks for your reply. I can just reset the model at each minute for meaningful measurement. @hello-binit Any other potential comment? The whole model is moving horizontally (not along the wheel direction), which is not the friction problem. I also removed one of the omni wheels and the model is still slipping.


Time 60s

Time 0s

rviz

@NateWright I am facing the same problem. Do you solve it? Thank you.

I messed with it for awhile, but I was unable to solve the issue

[mainakpal08]

I am facing the same issue. Any workaround ?

[cyb1212]

I changed the wheel control in stretch_gazebo_standard_gripper.urdf.xacro and the weight of the robot (make it heavy) to make the robot more stable. It is still slightly moving but if each step in your task is in 10 mins. It should be fine. Another supporting way is to use 'set_model_state' commend in gazebo before your observation.

I am facing the same issue. Any workaround ?

[mainakpal08]

I changed the wheel control in stretch_gazebo_standard_gripper.urdf.xacro and the weight of the robot (make it heavy) to make the robot more stable. It is still slightly moving but if each step in your task is in 10 mins. It should be fine. Another supporting way is to use 'set_model_state' commend in gazebo before your observation.

I am facing the same issue. Any workaround ?

Thanks @cyb1212 , it worked!

[tester23x]

I changed the wheel control in stretch_gazebo_standard_gripper.urdf.xacro and the weight of the robot (make it heavy) to make the robot more stable. It is still slightly moving but if each step in your task is in 10 mins. It should be fine. Another supporting way is to use 'set_model_state' commend in gazebo before your observation.

I am facing the same issue. Any workaround ?

@cyb1212 I have tried to change the weight of link_righ and left wheel mu2 value="2000.0". But still have this issue. This is really should not be the case at all. Please can the developer @hello-binit please resolve this important issue asap, as of now we cannot use stretch robot simulation for many things. Thank you.

[hello-binit]

@tester23x, could you help me reproduce this? You're using the latest commit and launching with roslaunch stretch_gazebo gazebo.launch? What behavior do you see? How long does it take for the issue to occur? Could you send a video of the issue you're seeing?

Technically, this repo is in maintenance mode, and we're recommending people switch to using Stretch Mujoco, which is a higher fidelity Mujoco simulation of Stretch. But I understand you may need functionality that only this repo can provide; I'll try to help when you've sent me your info for reproducing the issue.