def __init__(self,
reward_collision=-500,
reward_timestep=-5,
factor=100,
reward_target=500,
reward_target_collide=50,
repeat=2,
max_collision_len=4):
self.Rg = reward_target
self.Rc = reward_collision
self.Rt = reward_timestep
self.Reward_target_collide = reward_target_collide
self.factor = factor
self.repeat = repeat
self.done = False
self.queue = collections.deque(maxlen=repeat)
self.action_vel = [0.0, 0.6]
self.action_rotat = [-0.3, 0.0, 0.3] #kinda need to check
self.collision_queue = collections.deque(maxlen=repeat)
self.sim = Client()
self.action_space = []
self.target_collider_counter = 0
self.accuracy_limit = 0.6
for x in self.action_vel:
for y in self.action_rotat:
self.action_space.append([x, y])
self.action_space_n = len(self.action_space)
self.observation_space = [36]
def __init__( self ): self.entities = [] self.client = Client( 0.1, self.new_entity ) self.client.connect( "localhost", "1324", "1325", self.on_connection ) self.client.run()
class env_manager():
def __init__(self,
reward_collision=-500,
reward_timestep=-5,
factor=100,
reward_target=500,
reward_target_collide=50,
repeat=2,
max_collision_len=4):
self.Rg = reward_target
self.Rc = reward_collision
self.Rt = reward_timestep
self.Reward_target_collide = reward_target_collide
self.factor = factor
self.repeat = repeat
self.done = False
self.queue = collections.deque(maxlen=repeat)
self.action_vel = [0.0, 0.6]
self.action_rotat = [-0.3, 0.0, 0.3] #kinda need to check
self.collision_queue = collections.deque(maxlen=repeat)
self.sim = Client()
self.action_space = []
self.target_collider_counter = 0
self.accuracy_limit = 0.6
for x in self.action_vel:
for y in self.action_rotat:
self.action_space.append([x, y])
self.action_space_n = len(self.action_space)
self.observation_space = [36]
def _call_sim(self, action, flag):
sensor_data, velocity, angular_velocity, target_pos, collision = self.sim.step(
action, flag)
velocity = np.round_(math.sqrt(velocity[0]**2 + velocity[1]**2), 3)
speed = []
speed.append(velocity)
speed.append(angular_velocity)
speed = np.array(speed)
val = [[round(np.hypot(target_pos[0], target_pos[1]), 3)]]
state = np.concatenate((sensor_data.reshape(13,
1), speed.reshape(2, 1),
target_pos.reshape(2, 1), val))
return state, collision
def reset(self):
self.queue.clear()
self.done = False
state, collision = self._call_sim([0, 0], True)
for _ in range(self.repeat):
self.queue.append(state)
self.collision_queue.append(collision)
return np.array(self.queue).reshape(-1)
def step(self, action):
action_ = self.action_space[action]
state, collision = self._call_sim(action_, False)
self.queue.append(state)
self.collision_queue.append(collision)
reward = self._reward()
return np.array(self.queue).reshape(-1), reward, self.done
def _reward(self):
reward = 0.0
#target
val_1 = math.hypot(self.queue[1][15], self.queue[1][16]) - math.hypot(
self.queue[0][15], self.queue[0][16])
if (math.hypot(self.queue[0][15], self.queue[0][16]) <
self.accuracy_limit):
reward += self.Rg
self._academy()
self.done = True
else:
reward += self.factor * (
math.hypot(self.queue[1][15], self.queue[1][16]) -
math.hypot(self.queue[0][15], self.queue[0][16]))
#collision
if (self.collision_queue[0][0]):
reward += self.Rc
#target_collider
# if(self.collision_queue[0][1]):
# reward += self.Reward_target_collide
# self.done = True
#reward per time step
reward += self.Rt
if (self.collision_queue[1][0] and self.collision_queue[0][0]):
self.done = True
# self._academy()
return round(reward, 5)
def _academy(self):
self.target_collider_counter += 1
if self.target_collider_counter < 20:
self.accuracy_limit = self.accuracy_limit - 0.025 if self.accuracy_limit > 0.35 else 0.35
x = np.random.uniform(low=+1.0650e+00, high=+1.8910e+00)
y = np.random.uniform(low=-3.9050e+00, high=-3.2680e+00)
self.sim.target_position = [x, y, 0.0]
else:
if self.target_collider_counter == 21:
self.accuracy_limit = 0.525
self.sim.position = [+2.6500e+00, -4.3400e+00, +8.0299e-02]
self.accuracy_limit = self.accuracy_limit - 0.025 if self.accuracy_limit > 0.35 else 0.35
y = np.random.uniform(low=-3.5650e+00, high=-3.2180e+00)
x = np.random.uniform(low=+3.5360e+00, high=+4.3470e+00)
self.sim.target_position = [x, y, 0.0]
def end(self):
self.sim.end()
class ClientEntitySystem: def __init__( self ): self.entities = [] self.client = Client( 0.1, self.new_entity ) self.client.connect( "localhost", "1324", "1325", self.on_connection ) self.client.run() def on_connection( self, player_id ): print "Connected", player_id return self.connection def new_entity( self, data ): l, = struct.unpack( "!H", data[:2] ) cls = data[2: 2+l] print "new_ent", cls if cls == "BlockWorldState": w = World() self.append( w ) state = w.state() elif cls == "CharacterState": c = Character( (0., .1, 0.), (0., 0., 0.) ) w = TestWeapon( TestProjectile ) c.equip( w ) state = c.state() else: print "Unknown entity type" raise ValueError( "Unknown entity type" ) if state: data = state.deserialize( data[2+l:] ) return NewEntityData( data, state ) def append( self, ent ): self.entities.append( ent ) def draw( self ): for i in filter( lambda x: x.drawable, self.entities ): i.draw() batchable = list( filter( lambda x: x.batchable, self.entities ) ) batches = defaultdict( list ) for b in batchable: batches[ b.__class__ ].append( b ) for b in batches.values(): Batch( b, False ).draw() def update( self, dt ): self.input_entity = \ filter( lambda x: x.client_sendable, self.entities )[0] self.client.input( self.input_entity.state() ) for i in filter( lambda x: x.updatable, self.entities ): i.update( dt ) colls = list( filter( lambda x: x.collidable, self.entities ) ) for i, e in enumerate( colls[:] ): for o in colls[:][i+1:]: self.collide( e, o ) def collide( self, e1, e2 ): cg1 = e1.collision_geometry() cg2 = e2.collision_geometry() coll = cg1.collide( cg2 ) if coll.collided: e1.resolve( e2, coll )