Traffic Turbo is a road based environment where the agent (top left corner) is trained to reach his home (bottom right corner). The training & Testing for one of the random environments can be seen here
The environment consists of the following elements
The end goal of the agent is to take up an optimal path so as to keep a high reward at the end of the episode. Any move is considered invalid if
This has been done using Pygame library that provides GUI components & animation capabilities for python projects.
The agent has been trained using Q Learning technique in Reinforcement learning for ~2.k episodes using random states as initialization point for each episode.
For playing around, weights for 2 environments have been trained till 2k episodes & stored in env_weights function. For trying, initialize the game_env object with '1' or 'final_v'
Read more on Medium