CS401 Homework 5 Reinforcement Learning-Solved

Grid Map Environment
 white: the start position red: the goal position green: the obstacle black: ground obstacle reward: -1 goal reward: 10 other reward: 0 over the bound: -5

Question1
Please simulate the Monte Carlo Reinforcemenet learning with Exploring Starts under the given grid map environment.

Pseudocode:

Experimental Demonstration:

Question2
Please simulate the Sarsa (on-policy TD control)) algorithm under the given grid map environment.

Pseudocode:
Experimental Demonstration:

Question3
Please simulate the Q_learning (Off-policy TD Control) algorithm under the given grid map environment.

Pseudocode:
Experimental Demonstration:

 

Note: The above demonstrations are only parts of the output policy. Normally, your solution should be different but moving to the goal is necessary.

Question4 - Extra Credit
Please add the heuristic reward on the grid map, such as the DWA reward, A star reward, or distanceto-goal based reward learned form the previous lectures, to achieve a regular policy as you expected.

Coding instruction
Install the intelligent robotics simulator

git clone -b edu https://github.com/hanruihua/intelligent-robot-simulator.git cd intelligent-robot-simulator pip install -e . 

Note1: Please confirm that this repository is under the edu branch. You can use git branch to check current branch. If it is not under the edu branch, you can use git checkout edu to change current branch to edu branch.

Note2: The pycharm reduces the functionality of Matplotlib, which may lead to the failure of saving the gif animation. You can follow this link to solve this problem

Note3: If you have installed this simulator, you can use git pull to fetch the code update.

Code for questions

There are multiple files for these questions in the source folder.

 question1_run.py: is the main program you should run for question1 question2_run.py: is the main program you should run for question2 question3_run.py: is the main program you should run for question3 reinforcement_learning.py: is the library to implement three reinforcement learning algorithms: Monte Carlo Exploring Starts, Sarsa, and Q-learning. You should fill in this file to complete these three algorithms for the questions.

  grid_map.py: is the file that defines the class about the grid map for you to use. You can add heuristic reward here for extra question4.   map_matrix.npy and reward_matrix.npy: define the map and the reward in each grid.

You should complement the parts between ---- in the file reinforcement_learning.py for question1-3, and the file grid_map.py for extra question4. You can set the parameter animation = True in question_run.py to generate the animation.