Code implementing the CORE-RL algorithm with DDPG, PPO, and TRPO. We have included code for three environments: Car-Following, CartPole, and TORCS Racecar Simulator. Each folder contain a README file with more information about running the algorithm. Results are output as a .mat file for Car Following and CartPole, which can be read into MATLAB. TORCS results are output as .log files because the simulator for distinct training runs must be run on individual sockets.
For information about the algorithm and other additional details, see the paper "Control Regularization for Reduced Variance Reinforcement Learning". Please contact rcheng@caltech.edu for any bugs/issues.