-
Notifications
You must be signed in to change notification settings - Fork 487
(Don't merge without review!) Allow to build Roboschool using regular Bullet library using BulletRobotics #62
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.
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
base: master
Are you sure you want to change the base?
Conversation
https://github.com/bulletphysics/bullet3 using the new Extras/BulletRobotics library. First try to connect to SHARED_MEMORY, if it fails use DIRECT mode.
|
@erwincoumans I am trying to get this running on the latest version of macOS, but I am running into this error: Complete output from command python setup.py egg_info: Command "python setup.py egg_info" failed with error code 1 |
|
Oh, just found this. Looks great! |
|
There are some issues we are tracking down. We are converting the Roboschool envs to pybullet in a compatible way so we can exchange trained weights. Some envs work (such as Ant), but cheetah and hopper doesn't yet. We are tracking this down now here: Also, I fixed the MJCF capsule conversion code for TinyRenderer, and this breaks Roboschool. I'll look into a fix that works for both, or otherwise create some flag that Roboschool can call (during loadMJCF) for backwards compatibility of the rendering. The capsule change is in this commit: |
|
Hello, I am following up on this request - it seems the idea is to stop using a separate fork of Bullet - that would be great because Bullet has a lot of contributors and is frequently updated. Is there any reason for using a Bullet fork and not the original Bullet? |
I added the Extras/BulletRobotics library so it is easier to build Roboschool out-of-the-box. Note that the self-collision changes are all in the Bullet repo, I don't think you need any custom modifications.
In addition, it contains all the fixes I had to make to run on latest Mac OSX (Sierra 10.12 and High Sierra 10.13).
Finally, this will first try to connect to SHARED_MEMORY, if it fails use DIRECT mode. This will let me easier debug thing outside of Roboschool: spawn a pybullet server in 'connect(p.GUI_SERVER) etc.
If/when you are happy, we could create some tagged release in Bullet, so you don't get surprising regressions.