In this project, the robot(agent) avoids static/dynamic obstacles in the environment.
The blue robot has no prior knowledge of the environment. It doesn't know where the obstacles are. It uses seven sensors(e.g. ultrasonic/laser) to detect its surroundings, and uses these measured information to make decisions(move forward, turn left/right).
There are two different goals:
- Avoid obstacles in the environment
- Navigate towards the goal while avoiding obstacles
In the neural network folder, there are two sub folders.
Obstacle avoidance folder contains files that run the agent under static/dynamic envrionments.
The agent in this environment has one goal which is to avoid obstacles.
Reach destination folder contains files that run the agent under static/dynamic environments.
The agent has two goals. i) Avoid obstacles, and ii) navigate towards the goal(marked green).
Before running the program, the path must be changed to include the neural network folder
ex) my path to the neural network folder:
sys.path.insert(0,'/Users/sshin/Desktop/github/RLRobotSimulation/neural network')
##Notes
-The state of the agent is represented depending on the sensor readings.
-Q learning is used together with the function approximation.
-Occupancy grid mapping is used to locate obstacles in the environment.
(However, this information is not currently used to help the agent make decisions)
##Requirements
To run the project, follwing requriements should be installed:
-pygame installation
-matplotlib installation
-numpy installation
-tensorflow installation
##What to see
1)Current episode
2)Epsilon value(decides wheter to choose best action or random action)
3)Seven sensor readings(in range between 0 to 20)
4)Expected reward of each actions
5)Distance and relative angle to the goal(used to represent the state of the agent)
6)Current reward
7)Buttons to control the environment
8)Probabiliy that shows if detected obstacle is static
##To run the agent in static environment $ python2.7 static test.py
##To run the agent in dynamic environment $ python2.7 dynamic test.py
