PyTorch implementation of Episodic Memory Actor-Critic (EMAC).
TD3 and DDPG architecture parameters were based on official TD3 implementation: link
For training run:
python train.py --policy EMAC --env Walker2d-v3 --k 2 --alpha 0.1 --beta 0.1 --max_timesteps 200000 --device cuda:0
Paper training curves can be found in curves directory as saved TensorBoard logs in json format. For producing results below run
bash scripts/Walker2d-v3/train_EMAC.sh
