I recently began developing a c++ drl framework with libtorch with a simplified interface to enable researchers take advantage of multi-threading to acquire more batches and also c++ computation speed. I have built PPO and A2C to utilize this feature and i am blown away by the results and efficiency of resource use compared to multiprocessing. I need to work with other drl experts or software engineers to really bring this to life completely, as i believe drl research easily is limited by your accessible to large computing power.