def reset(self, **kwargs):
"""
Agents are given random positions in the map, targets are given random positions near a random agent.
Return an observation state dict with agent ids (keys) that refer to their observation
"""
obs_dict = {}
init_pose = self.get_init_pose(**kwargs)
# Initialize agents
for ii in range(self.num_agents):
self.agents[ii].reset(init_pose['agents'][ii])
obs_dict[self.agents[ii].agent_id] = []
# Initialize targets and beliefs
for jj in range(self.num_targets):
self.belief_targets[jj].reset(init_state=np.concatenate(
(init_pose['belief_targets'][jj][:2], np.zeros(2))),
init_cov=self.target_init_cov)
self.targets[jj].reset(
np.concatenate(
(init_pose['targets'][jj][:2], self.target_init_vel)))
#For each agent calculate belief of all targets
for kk in range(self.num_agents):
r, alpha = util.relative_distance_polar(
self.belief_targets[jj].state[:2],
xy_base=self.agents[kk].state[:2],
theta_base=self.agents[kk].state[2])
logdetcov = np.log(LA.det(self.belief_targets[jj].cov))
obs_dict[self.agents[kk].agent_id].extend(
[r, alpha, 0.0, 0.0, logdetcov, 0.0])
for agent_id in obs_dict:
obs_dict[agent_id].extend([self.sensor_r, np.pi])
return obs_dict
def step(self, action_dict):
obs_dict = {}
reward_dict = {}
done_dict = {'__all__':False}
info_dict = {}
# Targets move (t -> t+1)
for n in range(self.nb_targets):
self.targets[n].update()
self.belief_targets[n].predict() # Belief state at t+1
# Agents move (t -> t+1) and observe the targets
for ii, agent_id in enumerate(action_dict):
obs_dict[self.agents[ii].agent_id] = []
reward_dict[self.agents[ii].agent_id] = []
done_dict[self.agents[ii].agent_id] = []
action_vw = self.action_map[action_dict[agent_id]]
# Locations of all targets and agents in order to maintain a margin between them
margin_pos = [t.state[:2] for t in self.targets[:self.nb_targets]]
for p, ids in enumerate(action_dict):
if agent_id != ids:
margin_pos.append(np.array(self.agents[p].state[:2]))
_ = self.agents[ii].update(action_vw, margin_pos)
# Target and map observations
observed = np.zeros(self.nb_targets, dtype=bool)
# obstacles_pt = map_utils.get_closest_obstacle(self.MAP, self.agents[ii].state)
# if obstacles_pt is None:
obstacles_pt = (self.sensor_r, np.pi)
# Update beliefs of targets
for jj in range(self.nb_targets):
# Observe
obs, z_t = self.observation(self.targets[jj], self.agents[ii])
observed[jj] = obs
if obs: # if observed, update the target belief.
self.belief_targets[jj].update(z_t, self.agents[ii].state)
r_b, alpha_b = util.relative_distance_polar(self.belief_targets[jj].state[:2],
xy_base=self.agents[ii].state[:2],
theta_base=self.agents[ii].state[-1])
r_dot_b, alpha_dot_b = util.relative_velocity_polar(
self.belief_targets[jj].state[:2],
self.belief_targets[jj].state[2:],
self.agents[ii].state[:2], self.agents[ii].state[-1],
action_vw[0], action_vw[1])
obs_dict[agent_id].append([r_b, alpha_b, r_dot_b, alpha_dot_b,
np.log(LA.det(self.belief_targets[jj].cov)),
float(obs), obstacles_pt[0], obstacles_pt[1]])
# Assign target
for agent_id in obs_dict:
obs_dict[agent_id] = np.asarray(obs_dict[agent_id])
obs_dict[agent_id] = obs_dict[agent_id][None,self.greedy_dict[agent_id]]
# Get all rewards after all agents and targets move (t -> t+1)
reward, done, mean_nlogdetcov = self.get_reward(obstacles_pt, observed, self.is_training)
reward_dict['__all__'], done_dict['__all__'], info_dict['mean_nlogdetcov'] = reward, done, mean_nlogdetcov
return obs_dict, reward_dict, done_dict, info_dict
def step(self, action_dict):
obs_dict = {}
reward_dict = {}
done_dict = {'__all__': False}
info_dict = {}
# Targets move (t -> t+1)
for n in range(self.num_targets):
self.targets[n].update()
# Agents move (t -> t+1) and observe the targets
for ii, agent_id in enumerate(action_dict):
obs_dict[self.agents[ii].agent_id] = []
reward_dict[self.agents[ii].agent_id] = []
done_dict[self.agents[ii].agent_id] = []
action_vw = self.action_map[action_dict[agent_id]]
_ = self.agents[ii].update(action_vw,
[t.state[:2] for t in self.targets])
observed = []
for jj in range(self.num_targets):
# Observe
obs = self.observation(self.targets[jj], self.agents[ii])
observed.append(obs[0])
self.belief_targets[jj].predict() # Belief state at t+1
if obs[0]: # if observed, update the target belief.
self.belief_targets[jj].update(obs[1],
self.agents[ii].state)
obstacles_pt = map_utils.get_closest_obstacle(
self.MAP, self.agents[ii].state)
if obstacles_pt is None:
obstacles_pt = (self.sensor_r, np.pi)
for kk in range(self.num_targets):
r_b, alpha_b = util.relative_distance_polar(
self.belief_targets[kk].state[:2],
xy_base=self.agents[ii].state[:2],
theta_base=self.agents[ii].state[-1])
r_dot_b, alpha_dot_b = util.relative_velocity_polar(
self.belief_targets[kk].state[:2],
self.belief_targets[kk].state[2:],
self.agents[ii].state[:2], self.agents[ii].state[-1],
action_vw[0], action_vw[1])
obs_dict[agent_id].extend([
r_b, alpha_b, r_dot_b, alpha_dot_b,
np.log(LA.det(self.belief_targets[kk].cov)),
float(observed[kk])
])
obs_dict[agent_id].extend([obstacles_pt[0], obstacles_pt[1]])
# Get all rewards after all agents and targets move (t -> t+1)
reward, done, mean_nlogdetcov = self.get_reward(
obstacles_pt, observed, self.is_training)
reward_dict['__all__'], done_dict['__all__'], info_dict[
'mean_nlogdetcov'] = reward, done, mean_nlogdetcov
return obs_dict, reward_dict, done_dict, info_dict
def reset(self, **kwargs):
"""
Random initialization a number of agents and targets at the reset of the env epsiode.
Agents are given random positions in the map, targets are given random positions near a random agent.
Return an observation state dict with agent ids (keys) that refer to their observation
"""
try:
self.nb_agents = kwargs['nb_agents']
self.nb_targets = kwargs['nb_targets']
except:
self.nb_agents = np.random.random_integers(1, self.num_agents)
self.nb_targets = np.random.random_integers(1, self.num_targets)
obs_dict = {}
init_pose = self.get_init_pose(**kwargs)
# Initialize agents
for ii in range(self.nb_agents):
self.agents[ii].reset(init_pose['agents'][ii])
obs_dict[self.agents[ii].agent_id] = []
# Initialize targets and beliefs
for nn in range(self.nb_targets):
self.belief_targets[nn].reset(init_state=np.concatenate(
(init_pose['belief_targets'][nn][:2], np.zeros(2))),
init_cov=self.target_init_cov)
self.targets[nn].reset(
np.concatenate(
(init_pose['targets'][nn][:2], self.target_init_vel)))
# For nb agents calculate belief of targets assigned
for jj in range(self.nb_targets):
for kk in range(self.nb_agents):
r, alpha = util.relative_distance_polar(
self.belief_targets[jj].state[:2],
xy_base=self.agents[kk].state[:2],
theta_base=self.agents[kk].state[2])
logdetcov = np.log(LA.det(self.belief_targets[jj].cov))
obs_dict[self.agents[kk].agent_id].append(
[r, alpha, 0.0, 0.0, logdetcov, 0.0, self.sensor_r, np.pi])
# Greedily assign agents to closest target in order, all targets assigned if agents > targets
mask = np.ones(self.nb_targets, bool)
if self.nb_targets > self.nb_agents:
oracle = 1
else:
oracle = 0
for agent_id in obs_dict:
obs_dict[agent_id] = np.asarray(obs_dict[agent_id])
if np.sum(mask) != np.maximum(
0, self.nb_targets - self.nb_agents + oracle):
idx = np.flatnonzero(mask)
close = idx[np.argmin(obs_dict[agent_id][:, 0][mask])]
obs_dict[agent_id] = obs_dict[agent_id][None, close]
mask[close] = False
return obs_dict
def reset(self, **kwargs):
"""
Random initialization a number of agents and targets at the reset of the env epsiode.
Agents are given random positions in the map, targets are given random positions near a random agent.
Return an observation state dict with agent ids (keys) that refer to their observation
"""
try:
self.nb_agents = kwargs['nb_agents']
self.nb_targets = kwargs['nb_targets']
except:
self.nb_agents = np.random.random_integers(1, self.num_agents)
self.nb_targets = np.random.random_integers(1, self.num_targets)
obs_dict = {}
init_pose = self.get_init_pose(**kwargs)
# Initialize agents
for ii in range(self.nb_agents):
self.agents[ii].reset(init_pose['agents'][ii])
obs_dict[self.agents[ii].agent_id] = []
# Initialize targets and beliefs
for nn in range(self.nb_targets):
self.belief_targets[nn].reset(init_state=np.concatenate(
(init_pose['belief_targets'][nn][:2], np.zeros(2))),
init_cov=self.target_init_cov)
self.targets[nn].reset(
np.concatenate(
(init_pose['targets'][nn][:2], self.target_init_vel)))
# For nb agents calculate belief of targets assigned
for jj in range(self.nb_targets):
for kk in range(self.nb_agents):
r, alpha = util.relative_distance_polar(
self.belief_targets[jj].state[:2],
xy_base=self.agents[kk].state[:2],
theta_base=self.agents[kk].state[2])
logdetcov = np.log(LA.det(self.belief_targets[jj].cov))
obs_dict[self.agents[kk].agent_id].append(
[r, alpha, 0.0, 0.0, logdetcov, 0.0, self.sensor_r, np.pi])
# Assign agents to closest k targets, if less targets than k, consider all targets
self.k = 8
if self.nb_targets > self.k:
for agent_id in obs_dict:
obs_dict[agent_id] = np.asarray(obs_dict[agent_id])
close = np.argpartition(obs_dict[agent_id][:, 0], self.k)
obs_dict[agent_id] = obs_dict[agent_id][close[:self.k]]
else:
for agent_id in obs_dict:
obs_dict[agent_id] = np.asarray(obs_dict[agent_id])
return obs_dict
def observation(self, target, agent):
r, alpha = util.relative_distance_polar(target.state[:2],
xy_base=agent.state[:2],
theta_base=agent.state[2])
observed = (r <= self.sensor_r) \
& (abs(alpha) <= self.fov/2/180*np.pi) \
& (not(map_utils.is_blocked(self.MAP, agent.state, target.state)))
z = None
if observed:
z = np.array([r, alpha])
# z += np.random.multivariate_normal(np.zeros(2,), self.observation_noise(z))
z += self.np_random.multivariate_normal(np.zeros(2,), self.observation_noise(z))
'''For some reason, self.np_random is needed only here instead of np.random in order for the
RNG seed to work, if used in the gen_rand_pose functions RNG seed will NOT work '''
return observed, z
def update(self, observed, z_t, x_t, u_t=None):
# Kalman Filter Update
self.ukf.predict(u=u_t)
if observed:
r_pred, alpha_pred = util.relative_distance_polar(
self.ukf.x[:2], x_t[:2], x_t[2])
self.ukf.update(z_t,
R=self.obs_noise_func((r_pred, alpha_pred)),
agent_state=x_t)
cov_new = self.ukf.P
state_new = self.ukf.x
if LA.det(cov_new) < 1e6:
self.cov = cov_new
if not (self.collision_func(state_new[:2])):
self.state = np.clip(state_new, self.limit[0], self.limit[1])
def update(self, z_t, x_t):
"""
Parameters
--------
z_t : observation - radial and angular distances from the agent.
x_t : agent state (x, y, orientation) in the global frame.
"""
# Kalman Filter Update
r_pred, alpha_pred = util.relative_distance_polar(
self.state[:2], x_t[:2], x_t[2])
diff_pred = self.state[:2] - x_t[:2]
if self.dim == 2:
Hmat = np.array([[diff_pred[0], diff_pred[1]],
[-diff_pred[1] / r_pred, diff_pred[0] / r_pred]
]) / r_pred
elif self.dim == 4:
Hmat = np.array([
[diff_pred[0], diff_pred[1], 0.0, 0.0],
[-diff_pred[1] / r_pred, diff_pred[0] / r_pred, 0.0, 0.0]
]) / r_pred
else:
raise ValueError('target dimension for KF must be either 2 or 4')
innov = z_t - np.array([r_pred, alpha_pred])
innov[1] = util.wrap_around(innov[1])
R = np.matmul(np.matmul(Hmat, self.cov), Hmat.T) \
+ self.obs_noise_func((r_pred, alpha_pred))
K = np.matmul(np.matmul(self.cov, Hmat.T), LA.inv(R))
C = np.eye(self.dim) - np.matmul(K, Hmat)
cov_new = np.matmul(C, self.cov)
state_new = self.state + np.matmul(K, innov)
if True: #LA.det(cov_new) < 1e6:
self.cov = cov_new
self.state = np.clip(state_new, self.limit[0], self.limit[1])
def step(self, action_dict):
obs_dict = {}
reward_dict = {}
done_dict = {'__all__': False}
info_dict = {}
# Targets move (t -> t+1)
for n in range(self.nb_targets):
self.targets[n].update()
self.belief_targets[n].predict() # Belief state at t+1
# Agents move (t -> t+1) and observe the targets
for ii, agent_id in enumerate(action_dict):
obs_dict[self.agents[ii].agent_id] = []
reward_dict[self.agents[ii].agent_id] = []
done_dict[self.agents[ii].agent_id] = []
action_vw = self.action_map[action_dict[agent_id]]
# Locations of all targets and agents in order to maintain a margin between them
margin_pos = [t.state[:2] for t in self.targets[:self.nb_targets]]
for p, ids in enumerate(action_dict):
if agent_id != ids:
margin_pos.append(np.array(self.agents[p].state[:2]))
_ = self.agents[ii].update(action_vw, margin_pos)
# Target and map observations
observed = np.zeros(self.nb_targets, dtype=bool)
# obstacles_pt = map_utils.get_closest_obstacle(self.MAP, self.agents[ii].state)
# if obstacles_pt is None:
obstacles_pt = (self.sensor_r, np.pi)
# Update beliefs of targets
for jj in range(self.nb_targets):
# Observe
obs, z_t = self.observation(self.targets[jj], self.agents[ii])
observed[jj] = obs
if obs: # if observed, update the target belief.
self.belief_targets[jj].update(z_t, self.agents[ii].state)
r_b, alpha_b = util.relative_distance_polar(
self.targets[jj].state[:2],
xy_base=self.agents[ii].state[:2],
theta_base=self.agents[ii].state[-1])
r_dot_b, alpha_dot_b = util.relative_velocity_polar(
self.targets[jj].state[:2], self.targets[jj].state[2:],
self.agents[ii].state[:2], self.agents[ii].state[-1],
action_vw[0], action_vw[1])
obs_dict[agent_id].append(
[r_b, alpha_b, r_dot_b, alpha_dot_b,
float(obs)])
else:
# if no obs, take old obs + kalman like prediction (random noise)
# self._obs_dict[agent_id][jj][:2] += self.np_random.multivariate_normal(np.zeros(2,),
# self.observation_noise(self._obs_dict[agent_id][jj][:2]))
# self._obs_dict[agent_id][jj][2:4] += self.np_random.multivariate_normal(np.zeros(2,),
# self.observation_noise(self._obs_dict[agent_id][jj][:2]))
# self._obs_dict[agent_id][jj][4] = float(obs)
# obs_dict[agent_id].append(self._obs_dict[agent_id][jj])
obs_dict[agent_id].append([0.0, 0.0, 0.0, 0.0, float(obs)])
obs_dict[agent_id] = np.asarray(obs_dict[agent_id])
# store obs dict before shuffle
self._obs_dict[agent_id] = obs_dict[agent_id]
# shuffle obs to promote permutation invariance
self.rng.shuffle(obs_dict[agent_id])
# Get all rewards after all agents and targets move (t -> t+1)
reward, done, mean_nlogdetcov = self.get_reward(
obstacles_pt, observed, self.is_training)
reward_dict['__all__'], done_dict['__all__'], info_dict[
'mean_nlogdetcov'] = reward, done, mean_nlogdetcov
return obs_dict, reward_dict, done_dict, info_dict
