Veins-Gym exports Veins simulations as Open AI Gyms. This enables the application of Reinforcement Learning algorithms to solve problems in the VANET domain, in particular popular frameworks such as Tensorflow or PyTorch.
To install, simply run
pip install veins-gym (Veins-Gym on PyPI).
This project is licensed under the terms of the GNU General Public License 2.0.
Here are some frequent issues with Veins-Gym and its environments.
Please also checkout the guide on how to build new environments in doc/getting_started.md.
Sometimes the C++ code files generated by
protoc get out of sync with the installed version of
This leads to errors at link-time of the environment (not veins-gym itself, which does not contain C++-code) like:
undefined reference to google::protobuf::internal::ParseContext::ParseMessage
.pb.h files in the environment.
libprotocon your system (package names may differ based on the Linux distribution)
src/protobuf/veinsgym.pb.ccfiles (or however they may be called in the environment)
snakemake src/protobuf/veinsgym.pb.cc src/protobuf/veinsgym.pb.h(adapt for the files in your environment.
Also see Issue #1 of serpentine-env.
Normally, Veins-Gym swallows the standard output of veins simulations started by its environments. This reduces output clutter but makes debugging harder as error messages are not visible.
Enable veins standard output by adjusting
gym.register when running your agent:
"print_veins_stdout": True, # enable (debug) output of veins
Sometimes you want to run Veins simulations separately, e.g., with custom parameters, in a (different) container, or within a debugger. This is hard to achieve and less flexible when Veins-Gym launches the Veins simulation, as it does by default.
Solution: Disable auto-start of Veins in your agent's environment:
"run_veins": False, # do not start veins through Veins-Gym
"port": 5555, # pick a port to use
Then run your veins simulation manually, typically from a
scenario directory, in which the
omnetpp.ini file is located:
./run -u Cmdenv '--*.gym-connection.port=5555'
Make sure to use the same port there. Also, consider starting Veins first in order to avoid timeouts.
Veins-Gym expects a request from the Veins simulation after a certain time after
env.reset (which may in turn have started the Veins simulation).
If the Veins simulation takes a longer amount of (wall-clock) time to reach the point where the request is sent, the agent's environment may have timed out.
This is intended to notify about stuck simulations, but may not work as intended if the simulation is more complex or the host' hardware is less powerful than expected.
Solution: Increase the timeout when registering the gym environment.
"timeout": 10.0, # new timeout value (in seconds)
Thank for you job, very nice. Did it support simulation of C-V2X (LTE or NR)
Telecommunication Networks Group (TKN) at the School of Electrical Engineering and Computer Science, TU BerlinGitHub Repository Homepage
veins omnet sumo vehicular-networks simulator openai-gym car2x v2x vanet reinforcement-learning reinforcement-learning-environments