Mountain Car using Deep RL

Environment Details

A car is on a one-dimensional track, positioned between two "mountains". The goal is to drive up the mountain on the right; however, the car's engine is not strong enough to scale the mountain in a single pass. Therefore, the only way to succeed is to drive back and forth to build up momentum.

Solved Requirements

MountainCar-v0 defines "solving" as getting average reward of -110.0 over 100 consecutive trials.

The target_dqn.h5 model gets an average score of -99.67 in 100 consecutive episodes ans thus solves the Mountain Car environment.

Performance Of Trained Model

After 750 episodes of training

Files

1. target_dqn.h5 : A trained model for predicting Q values for state action pairs.
2. train.py : Contains detailed implementation of the DDQN for solving Mountain Car.
3. q_network.py : Contains the neural network architechture for predicting Q values.
4. test.py : Runs the trained model on the environment for 100 episodes and returns the mean score.

Testing

Epoch :88 Score :-104
Epoch :89 Score :-131
Epoch :90 Score :-104
Epoch :91 Score :-104
Epoch :92 Score :-88
Epoch :93 Score :-130
Epoch :94 Score :-103
Epoch :95 Score :-85
Epoch :96 Score :-104
Epoch :97 Score :-104
Epoch :98 Score :-102
Epoch :99 Score :-129

Average Reward  in 100 games : -100.58