Tags

DQN agent learns to interact with bridge to reach goal. Gym is force-based. Gym is no longer episodic - instead when reaching the goal, it flips north-south.

Also, many results added about experiments with ANN agent based on a O'reilly URL.
Base point for implementing multi agent interaction and test emergent tool use

Ending experiments with ANN and policy gradients. Final stage before going back to DQN

Added town gym with force-based actions and Policy Gradient trained agent, with complex architecture, from Simonini Thomas course.

Good base point to start creating a non-toy gym, probably using a Physics simulator or a game engine

Snake minigame-style with fixed goal, fixed initial position. Simple Q learning agent.

It's basically a has-less-bugs version of the 0.0.1, and also reaches the optimum solution at around 5800 iterations

Snake minigame-style with fixed goal, random initial position per episode OpenAI gym. Single Q learning agent.