The objective is to get an under powered car to the top of a hill (top = 0.5 position)
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.
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 TrueTo run with pre-trained weights:
python dueling_network.py testReward Plot:
The obtained result:
