Q-learning is a reinforcement learning technique used in machine learning. The goal of Q-learning is to learn a policy, which tells an agent what action to take under what circumstances. It does not require a model of the environment and can handle problems with stochastic transitions and rewards, without requiring adaptations.
In this tool, we are using curses to render our map and various simulation statistics such as Q-Table entry at a particular state. The following are some examples of the program running:
| Untrained Agent Exploration | Trained Agent Navigation |
|---|---|
 |
 |
You can build your own map as long as it fits into a rectangular shape. Alternatively, you can experiment with the few maps included in this repository. Start the simulation with the following commands:
python rlearn.py <name of map> Example : python rlearn.py cliff.txt
You can press space to skip a few training generations and esc to exit the simulation.
Apart from the map, you may feel free to modify various aspects of the code such as refresh rate of the visualization, if you wish to have pre-training and the learning policy.
Learning Policy - In the initialization of the Agent object, you can state if you wish to use qlearn (default) or sarsa policies
Refresh Rate - Line 290 : sleep_time = 0.1, modify the corresponding per-loop sleep time
Pre-Training - Line 294 : my_agent.fastforward(100000), modify the amount of cycles to move forward
When creating your own map, the following are the interpreted symbols:
'x' - represents a deadly object that the agent should avoid stepping on
'.' - represents an obstacle that is non-lethal, but blocks any movement
'S' - represents the starting point
'G' - represents the end point