In this project, we are interested in implementing Q-learning as an extension to PA3. Similar to MDP, we want to find an optimal policy for each grid in a given map. However, this time probabilistic models are not known and have to be learned. This can be a common case where robot has no idea how accurate its movements are and Q-learning ensures the optimization of the final policy list even when robot takes undesired movements in some iterations.
python learning.py
by default, the algorithm will run with configuration.json. To apply the other configuration(s), simply rename.
*For the sake of clarity and simplicity, we did not use ROS for simulation. *