远程服务器运行CoppeliaSim

For those asynchronous DRL algorithm, ‘PS-Workers’ is one of common framework with TensorFlow implementation, in which each worker often need run on one of GPU of remote server or cluster. Of cause, we can set the num_workers = 1 to run with PC, but in this manner it will greatly slow down the period of overall training, even reach a local maximumn with oscillation. In next, we post how to setup and run CoppeliaSim on remote server (Centos 7 platform).

Although, CoppeliaSim only release formal package on Ubuntu platform. We have proven that it can also run on Centos7 platform. Therefore, we suggest you download CoppeliaSim (Ubuntu 16.04) and copy it to remote server.

And add following command to ~/.bashrc:

export DISPLAY=:0.0
export COPPELIASIM_ROOT=[The absoulute path of your CoppeliaSim directory in remote server]
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$COPPELIASIM_ROOT
export QT_QPA_PLATFORM_PLUGIN_PATH=$COPPELIASIM_ROOT

Because CoppeliaSim relies on GLIBCXX 3.4.21, you should first check the GLIBCXX version by following cammand:

$: strings /usr/lib64/libstdc++.so.6 | grep GLIBCXX

If strings results do not include 3.4.21, we suggest you to compile higher version GCC with source code (GCC 7 or 8). And then, copy new libstdc++.so.6.0.* file to /usr/lib64 and create new soft link named as ‘libstdc++.so.6’ that direct to /usr/lib64/libstdc++.so.6.0.*.

once you have updated and setupped GCC, you can run previous command to check GLIBCXX version.

Download virtualgl rpm package and install it.

$: cd CoppeliaSim_PATH
$: ./coppeliaSim -h &