rl.deep_q_learning.DeepQLearner Class Reference

The Deep Q learner class. More...

Public Member Functions
None	__init__ (self, str env_path, str urdf_path=None, str urdf=None, bool use_graphics=False, str network_path="")
	The DeepQLearner class initializer. More...
def	handler (self, *_)
	Handler for saving the Deep Q-network upon interrupting the execution of the program.
None	save (self, str path="./src/rl/networks/most_recently_saved_network.pkl")
	Save the trained network in a pickle file. More...
np.ndarray	get_action_space (self, number_of_joints)
	Get all possible actions given the amount of joints. More...
Tuple[int, np.ndarray, Tuple]	get_random_wall (self)
	Get one of the possible walls at random. More...
DQN	make_dqn (self, network_path="")
	Create an instance of the DQN. More...
Tuple[int, np.ndarray]	step (self, np.ndarray state)
	Move 1 step forward in the simulation. More...
float	learn (self, int number_of_episodes=10000, int steps_per_episode=1000, bool logging=False)
	The training or testing loop (depending on variable 'training') of the deep q-learning. More...
int	predict (self, np.ndarray state)
	Pick the best or a random action depending on epsilon value of the Deep-Q Network. More...

Public Attributes
	goal_ball_diameter
	use_walls
	env
	x_range
	y_range
	z_range
	actions
	dqn
	training
	logger
	walls
	wall_centers
	current_wall_index

Static Public Attributes
int	STATE_SIZE = 6
int	WALL_STATE_SIZE = 9

Detailed Description

The Deep Q learner class.

Defines the class that will learn based on a Deep-Q Network.

Constructor & Destructor Documentation

__init__()

None rl.deep_q_learning.DeepQLearner.__init__	(		self,
		str	env_path,
		str	urdf_path = None,
		str	urdf = None,
		bool	use_graphics = False,
		str	network_path = ""
	)

The DeepQLearner class initializer.

Parameters

env_path	Path of the environment executable.
urdf_path	Path to the robot urdf file.
urdf	Instance that represents the robot urdf.
use_graphics	Boolean that turns graphics on or off.
network_path	Path to the Deep-Q network that should be used.

Returns

An instance of the DeepQLearner class.

Member Function Documentation

get_action_space()

np.ndarray rl.deep_q_learning.DeepQLearner.get_action_space	(		self,
			number_of_joints
	)

Get all possible actions given the amount of joints.

Parameters

number_of_joints

The amount of joints the robot arm has. A complex module has 2 joints, rotating and tilting.

Returns

A list of the actions

get_random_wall()

Tuple[int, np.ndarray, Tuple] rl.deep_q_learning.DeepQLearner.get_random_wall

(

self

)

Get one of the possible walls at random.

Returns

A wall.

learn()

float rl.deep_q_learning.DeepQLearner.learn	(		self,
		int	number_of_episodes = 10000,
		int	steps_per_episode = 1000,
		bool	logging = False
	)

The training or testing loop (depending on variable 'training') of the deep q-learning.

Parameters

number_of_episodes	Maximum amount of episodes.
steps_per_episode	Maximum amount of steps each episode.
logging	If true there will be wandb logs

Returns

Success rate throughout training.

make_dqn()

DQN rl.deep_q_learning.DeepQLearner.make_dqn	(		self,
			network_path = ""
	)

Create an instance of the DQN.

Parameters

network_path

Path to network, if the goal is testing, not training.

Returns

A DQN instance.

predict()

int rl.deep_q_learning.DeepQLearner.predict	(		self,
		np.ndarray	state
	)

Pick the best or a random action depending on epsilon value of the Deep-Q Network.

Parameters

state

Current state of the environment.

Returns

The chosen action.

save()

None rl.deep_q_learning.DeepQLearner.save	(		self,
		str	path = "./src/rl/networks/most_recently_saved_network.pkl"
	)

Save the trained network in a pickle file.

Parameters

path	Path to file where the network will be saved.

step()

Tuple[int, np.ndarray] rl.deep_q_learning.DeepQLearner.step	(		self,
		np.ndarray	state
	)

Move 1 step forward in the simulation.

Parameters

state

Current state of the environment.

Returns

The action that was taken and the observations that were made.

---

The documentation for this class was generated from the following file:

• src/rl/deep_q_learning.py