The docker command that works for me on Linux (NixOS) is this:
docker run \
--device nvidia.com/gpu=all \
--network=host \
--shm-size=512m \
-e DISPLAY=$DISPLAY \
-v /tmp/.X11-unix:/tmp/.X11-unix \
-v $(pwd):/workspace \
-it \
--rm \
-w /workspace/agent \
vtprl:py310 bash
Note the --network=host. This means ip_address parameter can be set to 127.0.0.1 so the simulator listens on this address and the agent connects to this address.
If using dockers network, the agent would need to connect to host.docker.internal(to connect to the host machine), and the host would need to listen to 0.0.0.0 to allow connections from the docker network. This does not work with the current config implementation because the address is shared bx agent and simulator.
Also, the --gpus all had to be changed to --device nvidia.com/gpu=all.
The docker command that works for me on Linux (NixOS) is this:
Note the
--network=host. This meansip_addressparameter can be set to127.0.0.1so the simulator listens on this address and the agent connects to this address.If using dockers network, the agent would need to connect to
host.docker.internal(to connect to the host machine), and the host would need to listen to0.0.0.0to allow connections from the docker network. This does not work with the current config implementation because the address is shared bx agent and simulator.Also, the
--gpus allhad to be changed to--device nvidia.com/gpu=all.