Skip to content

Latest commit

 

History

History

mountain_car

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 

The objective is to get an under powered car to the top of a hill (top = 0.5 position)

Details

The agent only knows his position and velocity in evry step. The agent can take one action from three: push left, no push, and push right. In this example, dueling network agent was used.

Solving with Dueling Network

Following are the commands used to train and test the model:

To train the model:

python dueling_network.py train --itr 1000 --capacity 10000 --batch 80 --save True --plot True

To run with pre-trained weights:

python dueling_network.py test

Results

Reward Plot:

The obtained result: