32 projects in the framework of Deep Reinforcement Learning algorithms: Q-learning, DQN, PPO, DDPG, TD3, SAC, A2C and others. Each project is provided with a detailed training log.
Hi, I've been looking through your code as a reference to figure out how to solve CarRacing-v0.
Mine works up to a point then has a catastrophic performance crash.
The only difference I can find between my version and yours is that When the unwrapped environment is done (fails) the agent gets a big negative reward.
You removed this in your wrapper, and I don't understand why.
What's the significance of offsetting the reward there?
The figure and log in README shows scores >1000, which due to the CarRacing's design, is not quite possible.
It turns out that the reward shaping in Wrapper.step() is not removed in evaluation and that leads to incorrect results.
Commenting out relevant lines, I got an average score of 820 over 100 episodes.