def step(self):
self.actnum += 1
prev_state = self.state
# Update state
qdr, dist = qdrdist(traf.lat[self.acidx], traf.lon[self.acidx],
traf.ap.route[self.acidx].wplat[-2],
traf.ap.route[self.acidx].wplon[-2])
t = agent.sta - ETA(agent.acidx)
print('STA {}, ETA {}, t {}'.format(agent.sta, ETA(agent.acidx), t))
hdg_rel = degto180(qdr - traf.hdg[agent.acidx])
# self.state = np.array([dist, t, hdg_ref,
# traf.tas[agent.acidx]])
self.state = np.array([dist, t, hdg_rel / 180.])
# Check episode termination
if dist < 1 or agent.sta - sim.simt < -60:
if agent.epsilon > agent.epsilon_min:
agent.epsilon -= 0.9 / 1000.
self.done = True
env.reset()
reward = self.gen_reward()
print('State {}'.format(self.state))
print('Reward {}, epsilon {}'.format(reward, agent.epsilon))
if train_phase:
self.log()
return self.state, reward, self.done, prev_state
def test():
eventmanager.update()
for i in eventmanager.events:
if env.actnum == 0:
agent.sta = ETA(agent.acidx) + random.random() * 100
# print('STA ', agent.sta)
next_state, reward, done, prev_state = env.step()
if not done:
agent.act_test(next_state)
def train():
eventmanager.update()
for i in eventmanager.events:
if env.actnum == 0:
agent.sta = ETA(agent.acidx) + random.random() * 100
next_state, reward, done, prev_state = env.step()
next_state = np.reshape(next_state, [1, agent.state_size])
if env.actnum > 0:
agent.remember(prev_state, agent.action, reward, next_state, done)
if len(agent.memory) > agent.batch_size:
agent.replay(agent.batch_size)
if not done:
agent.act(next_state)
def test():
eventmanager.update()
# print('ETA {} wp {}'.format(ETA(agent.acidx), traf.ap.route[0].wpname[traf.ap.route[0].iactwp]))
for i in eventmanager.events:
if env.actnum == 0:
agent.sta = ETA(agent.acidx) + random.random() * 100
# print('STA ', agent.sta)
next_state, reward, done, prev_state = env.step()
if not done:
agent.act_test(next_state)
f = open(agent.testname, 'a')
f.write("{},{},{},{},{},{},{},{},{},{}\n".format(env.ep, env.actnum, env.reward, env.state[0], env.state[1], env.state[2], agent.sta, sim.simt, traf.lat[env.acidx], traf.lon[env.acidx]))
f.close()
if env.ep>25:
sim.stop()