def reset(self):
""" re-initializes the environment, setting the cart back in a random position.
"""
Environment.reset(self)
angle = random.uniform(-0.2, 0.2)
pos = random.uniform(-1., 1.)
self.sensors = (angle, 0.0, pos, 0.0)
def getState(self):
""" returns the state one step (dt) ahead in the future. stores the state in
self.sensors because it is needed for the next calculation. The sensor return
vector has 4 elements: theta, theta', s, s' (s being the distance from the
origin).
"""
Environment.getState(self)
return asarray(self.sensors).flatten()
def performAction(self, action):
""" stores the desired action for the next runge-kutta step.
'actionNum' discrete actions are available, which are mapped
to the discrete even-spaced interval of [-50, 50] Newton.
"""
action = array([((2.0*action[0] / (float(self.conditions['actionNum'])-1.)) - 1.) * 50.])
Environment.performAction(self, action)
def __init__(self, topology, goal):
Environment.__init__(self)
self.mazeTable = topology
self.goal = goal
if self.initPos == None:
self.initPos = self._freePos()
self.initPos.remove(self.goal)
self.reset()
def __init__(self, maxSteps=100):
Environment.__init__(self)
# initialize the environment (randomly)
self.reset()
self.action = 0.0
self.maxSteps = maxSteps
self.centerCart = False
self.generator = True
def getReward(self):
Environment.getReward(self)
angle = abs(self.sensors[0])
s = abs(self.sensors[2])
reward = 0
if angle < 0.05 and (s < 0.05 or not self.centerCart):
reward = +2.
elif angle > 0.7 or s > 2.4:
reward = 0.
else:
if self.centerCart:
reward = max(+1., 2*(1.-s))
else:
reward = +1.
return reward
def reset(self):
""" return to initial position (stochastically): """
Environment.reset(self)
self.bang = False
self.perseus = choice(self.initPos)
self.state = self._lookAround()
def reset(self):
Environment.reset(self)
self.state = np.random.uniform(3, 4)
def episodeFinished(self):
Environment.episodeFinished(self)
return self.timestep >= 1
def __init__(self):
Environment.__init__(self)
self.reset()
def reset(self):
Environment.reset(self)
self.state = random.uniform(-1, 1, 1)
self.target = array([0.4])
def resetToState(self, state):
Environment.reset(self)
self.sensors = state
def performAction(self, action):
""" stores the desired action for the next runge-kutta step. Actions are expected
to be between -50 and 50.
"""
action = clip(asarray(action), -50., 50.)
Environment.performAction(self, action)
