This project is for the BLG604E Deep Reinforcement Learning term project at ITU, taught by Nazım Kemal Üre.
Competition took place within TORCS environment and the winner agent was our Pesla agent. There is an upcoming conference paper regarding architecture. Example races
Winner agent:
python test_torcs.py --test --algo sac-lstm --load-from releases/TORCS_SACLSTM_512256128_L1_EP5000_N1_G99.pt
Training: (please see command line arguments)
python run_torcs.py
There are multiple agents under releases folder. But their architecture is not all the same. I would update here for each of them in time.
This project and repository is the joint work of Kıvanç Güçkıran, Can Erhan and Onur Karadeli.
The codebase is a clone of and heavily borrowed from https://github.com/medipixel/rl_algorithms/