A simple RL-based on-ramp merging scenario simulator. Designed for two-player continuous Decision Making Cognitive Behavior Experiments.
Gym is needed:
pip install gym=0.20.0
merging_env in gym:
pip install -e .
To train models, set load_path to the model you want, then run the python script. For example:
load_path=None
upper = Goal_DQN(load_path)
lower = HDQN(load_path)
For testing, run "human_player.py". You can set
OP_MODEL="pvp"
so that 2 players are controlled by keyboard. w/s for P1 and Keypad 0/1/2/3/4 actions for P2.
A pretrained model is needed to play with agents. Copy the model files to the folder "scripts/", and set
load_path=name_of_the_model
in human_player.py. Don't forget to specify the network type (e.g. OP_MODEL = "dqn")