DriveML Huawei Autonomous Vehicles Challenge Forum

Go back to competition Back to thread list Post in this thread

> Has any of you been using completely non-python programming languages and non-RLLib frameworks to train the model in the openAI gym environment?

We've been attempting to deploy C++ codes for model training since a month ago, however, no matter how many different ways we tried to apply C++ codes in the policy, it results in a failure because we strictly followed each and every step instructed in the README.MD file.
We've tried manually modifying the policy file for python wrappers to make C++ codes work, we've tried SWIG for Ubuntu 64-bit edition, we've tried pybind11 for Ubuntu 64-bit edition, we've failed to deploy cython and boost::python. We were unable to add any C functions in the policy script.
We've checked the official site of openAI and the gym environment, but none of the information indicated any support for C++ and other non-python programming languages.
We've been wondering about whether the pathways to use other programming languages are 100% blocked up, as the original starter kit codes support the RLLib framework by the Ray team ONLY, which blocks up all possibilities to deploy other frameworks for all languages including python.
Has any of you been successfully generating models using non-python programming languages with frameworks apart from the Ray RLLib in the openAI gym environment? We are seeking for immediate clarification about this restriction issue on the framework and the programming languages for training our models.
Please reply informing only the languages and the framework you've tried on with. TIA.

Posted by: Team115 @ Dec. 10, 2019, 10:26 p.m.

Though I can't specifically comment on C/C++ support via Cython or CPython for Gym, or Ray - the online scoring program once you submit your application will not support these submission.

May I ask how you intend to use C/C++ for your solution?

Kindly,
Julian (HiWay Simulator Team)

Posted by: HuaweiUK @ Dec. 11, 2019, 5:42 a.m.

There are famous frameworks for C++ so we are attempting to implement those in place of the Ray RLLib in the openAI gym environment to train our models. Technically, we began with an easier version of wrapping C++ codes in our policy script file, but we found that all solutions to wrap C++ codes in our policy script file are dead blocked and result in alienated errors. Therefore, we are still attempting to deploy C++ codes with C++ frameworks in the openAI gym environment to swap out the Ray RLLib framework for potential improvements of performance of our models.

Posted by: Team115 @ Dec. 12, 2019, 1:35 p.m.
Post in this thread