Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

MjSim rendering doesn't use GPU #493

Closed
prasoongoyal opened this issue Dec 16, 2019 · 7 comments
Closed

MjSim rendering doesn't use GPU #493

prasoongoyal opened this issue Dec 16, 2019 · 7 comments

Comments

@prasoongoyal
Copy link

I'm running body_interaction.py example file on a remote server. Here's the code after making changes to run in headless mode:

#!/usr/bin/env python3
"""
Example of how bodies interact with each other. For a body to be able to
move it needs to have joints. In this example, the "robot" is a red ball
with X and Y slide joints (and a Z slide joint that isn't controlled).
On the floor, there's a cylinder with X and Y slide joints, so it can
be pushed around with the robot. There's also a box without joints. Since
the box doesn't have joints, it's fixed and can't be pushed around.
"""
from mujoco_py import load_model_from_xml, MjSim, MjViewer
import math
import os

MODEL_XML = """
<?xml version="1.0" ?>
<mujoco>
    <option timestep="0.005" />
    <worldbody>
        <body name="robot" pos="0 0 1.2">
            <joint axis="1 0 0" damping="0.1" name="slide0" pos="0 0 0" type="slide"/>
            <joint axis="0 1 0" damping="0.1" name="slide1" pos="0 0 0" type="slide"/>
            <joint axis="0 0 1" damping="1" name="slide2" pos="0 0 0" type="slide"/>
            <geom mass="1.0" pos="0 0 0" rgba="1 0 0 1" size="0.15" type="sphere"/>
			<camera euler="0 0 0" fovy="40" name="rgb" pos="0 0 2.5"></camera>
        </body>
        <body mocap="true" name="mocap" pos="0.5 0.5 0.5">
			<geom conaffinity="0" contype="0" pos="0 0 0" rgba="1.0 1.0 1.0 0.5" size="0.1 0.1 0.1" type="box"></geom>
			<geom conaffinity="0" contype="0" pos="0 0 0" rgba="1.0 1.0 1.0 0.5" size="0.2 0.2 0.05" type="box"></geom>
		</body>
        <body name="cylinder" pos="0.1 0.1 0.2">
            <geom mass="1" size="0.15 0.15" type="cylinder"/>
            <joint axis="1 0 0" name="cylinder:slidex" type="slide"/>
            <joint axis="0 1 0" name="cylinder:slidey" type="slide"/>
        </body>
        <body name="box" pos="-0.8 0 0.2">
            <geom mass="0.1" size="0.15 0.15 0.15" type="box"/>
        </body>
        <body name="floor" pos="0 0 0.025">
            <geom condim="3" size="1.0 1.0 0.02" rgba="0 1 0 1" type="box"/>
        </body>
    </worldbody>
    <actuator>
        <motor gear="2000.0" joint="slide0"/>
        <motor gear="2000.0" joint="slide1"/>
    </actuator>
</mujoco>
"""

model = load_model_from_xml(MODEL_XML)
sim = MjSim(model)
img = sim.render(600, 600, device_id=1)
t = 0
while True:
    sim.data.ctrl[0] = math.cos(t / 10.) * 0.01
    sim.data.ctrl[1] = math.sin(t / 10.) * 0.01
    t += 1
    sim.step()
    img = sim.render(600, 600, device_id=1)
    if t > 100 and os.getenv('TESTING') is not None:
        break

Even though I specify device_id=1 in sim.render(), the process doesn't use the GPU. Neither does it throw any errors; it just falls back to using the CPU for rendering.
Any suggestions?

Thank you!
@SudeepDasari
Copy link

Maybe you're having the same issue I had? Check out #512

@wookayin
Copy link
Contributor

Make sure mujoco_py is using a shared module built with GPU:

>>> import mujoco_py
>>> mujoco_py.cymj

If it prints something like mujoco_py/generated/cymj_2.0.2.9_37_linuxgpuextensionbuilder_37.so, you are good to go. Otherwise, you may need to rebuild it:

$ pip uninstall mujoco_py
$ pip install mujoco_py --no-binary mujoco_py    # force rebuilding shared modules

You need to have libopengl-dev, libglew-dev, etc. installed. Also, make sure that /usr/lib/nvidia-000 directory exists (you may need to create on your own) -- this is a bug of openai mujoco_py. Refer to: #408 (comment)

@prasoongoyal
Copy link
Author

Thanks for the suggestions! I'll try those out.
For now, I was able to get it work using a hack -- I installed dm_control and imported mujoco from dm_control:
from dm_control import mujoco
That fixed the issue.

@robotlearning123
Copy link

Make sure mujoco_py is using a shared module built with GPU:

>>> import mujoco_py
>>> mujoco_py.cymj

If it prints something like mujoco_py/generated/cymj_2.0.2.9_37_linuxgpuextensionbuilder_37.so, you are good to go. Otherwise, you may need to rebuild it:

$ pip uninstall mujoco_py
$ pip install mujoco_py --no-binary mujoco_py    # force rebuilding shared modules

You need to have libopengl-dev, libglew-dev, etc. installed. Also, make sure that /usr/lib/nvidia-000 directory exists (you may need to create on your own) -- this is a bug of openai mujoco_py. Refer to: #408 (comment)

Hi @wookayin. I got the output like this: mujoco_py/generated/cymj_2.0.2.13_37_linuxcpuextensionbuilder_37.so, and even I rebuild it, I cannot get the gpu version. Do you know how to deal with this?

@wenbin-hu
Copy link

Make sure mujoco_py is using a shared module built with GPU:

>>> import mujoco_py
>>> mujoco_py.cymj

If it prints something like mujoco_py/generated/cymj_2.0.2.9_37_linuxgpuextensionbuilder_37.so, you are good to go. Otherwise, you may need to rebuild it:

$ pip uninstall mujoco_py
$ pip install mujoco_py --no-binary mujoco_py    # force rebuilding shared modules

You need to have libopengl-dev, libglew-dev, etc. installed. Also, make sure that /usr/lib/nvidia-000 directory exists (you may need to create on your own) -- this is a bug of openai mujoco_py. Refer to: #408 (comment)

Hi @wookayin. I got the output like this: mujoco_py/generated/cymj_2.0.2.13_37_linuxcpuextensionbuilder_37.so, and even I rebuild it, I cannot get the gpu version. Do you know how to deal with this?

Hi @wangcongrobot , I had the same issue with you. Did you solve this problem?

@Amir-Arsalan
Copy link

@prasoongoyal How do you check whether or not MuJoCo is running stuff on the GPU?

@HareshKarnan
Copy link

HareshKarnan commented Apr 16, 2021 •
edited
Loading

^ you can check the output of nvidia-smi. It should show a "G" to the left of the python process.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
7 participants
@wookayin @HareshKarnan @Amir-Arsalan @robotlearning123 @prasoongoyal @SudeepDasari @wenbin-hu