mgui7 / AppliedRL_TUM

group-05

The project code of course "Applied Reinforcement Learning" in Technical University of Munich Summer Semester 2020.

Team members

• B.Sc. Yikai Kang (03728450)
• B.Sc. Ming Gui (03687866)
• B.Sc. Bowen Ma (03721259)

Goal of Project

In this Project, Reinforcement Learning is implemented to solve the task that robot base and robotic arms can get all scores in movement.

How to use

1. Open main.py
2. Run

Introduction of files and folders

Folders

• documentation: The figure of exponential reward and MATLAB code.
• code/maps: Path of robot is saved in maps.
• code/resources: Some image items which are read by Viewer class.
• code/misc: miscellaneous folder

Files

• code/quick_start.py: "Hello World" file and run it directly.
• code/main.py: Clean py file to quick testing. It will import classes and functions from env.py, therefore this file is very clean and you can test your algorithm directly.
• code/reward_calculate.py: We provided different reward modes to test the performance of RL algorithm.
• code/rl.py: Reinforcement learning algorithm.
• code/elements.py: Save some global variables.

Introduction of States

You may see 5 items in state list variable. For example, you can see the following numbers: [570.0, 200.00000000000165, 45, 45, (1, 1, 1)]

1. x position of robot base
2. y position
3. Angle of arm1
4. Angle of arm2
5. Status of scores: 1 means it still exists, 0 means it have been collected.

Introduction of Actions

Three different types of actions. The action variable is also a list. It looks like: [1,0,1]

1. Robot base direction: 1 or 0 or -1
2. Arm1 direction: 1 or 0 or -1
3. Arm2 direction: 1 or 0 or -1 1 means move(rotate) forwards, 0 means don't move and -1 means move(rotate) backwards.