def experiment(env_id, traj_id, verbose=False):
obs_list = get_obs('acrobot_obs', env_id)
data = load_data('acrobot_obs', env_id, traj_id)
ref_path = data['path']
start_goal = data['start_goal']
env_vox = torch.from_numpy(np.load('mpnet/sst_envs/acrobot_obs_env_vox.npy')[env_id]).unsqueeze(0).float()
tic = time.perf_counter()
final_start = ref_path[0]
final_goal = ref_path[-1]
## initiate planner
params = get_params(final_goal)
mpc_mpnet = MPCMPNetPlanner(
params,
final_start,
final_goal,
env_vox,
system="acrobot_obs",
setup="default_norm",
#setup="default_norm_aug",
#setup="norm_nodiff_noaug_20step2e-2",
ep=5000,
obs_list=obs_list[env_id],
verbose=verbose)
mpc_mpnet.mpnet.train()
## start experiment
it = 0
while it < params['max_plan_it'] and not mpc_mpnet.reached:
mpc_mpnet.reset()
for i in range(20):
if verbose:
print('iteration:', i)
mpc_mpnet.step()
if mpc_mpnet.reached:
break
it += 1
toc = time.perf_counter()
# print(mpc_mpnet.costs)
costs = np.sum(mpc_mpnet.costs) if len(mpc_mpnet.costs) > 0 else np.inf
result = {
'env_id': env_id,
'traj_id': traj_id,
'planning_time': toc-tic,
'successful': mpc_mpnet.reached,
'costs': costs
}
print("env {}, traj {}, {}, time: {} seconds, {}(ref:{}) costs".format(
env_id,
traj_id,
result['successful'],
result['planning_time'],
result['costs'],
data['cost'].sum()))
return result
def experiment(env_id,
traj_id,
verbose=False,
system='cartpole_obs',
params=None):
print("env {}, traj {}".format(env_id, traj_id))
# print(params)
if system in ['cartpole_obs', 'acrobot_obs', 'car_obs']:
obs_list = get_obs(system, env_id)[env_id].reshape(-1, 2)
elif system in ['quadrotor_obs']:
obs_list = get_obs_3d(system, env_id)[env_id].reshape(-1, 3)
data = load_data(system, env_id, traj_id)
ref_path = data['path']
start_goal = data['start_goal']
# print(start_goal)
env_vox = np.load('mpnet/sst_envs/data/{}_env_vox.npy'.format(system))
obc = env_vox[env_id]
# print(obc.reshape(-1), obc.reshape(-1).shape)
# print(obs_list)
number_of_iterations = params['number_of_iterations'] # 3000000#
min_time_steps = params[
'min_time_steps'] if 'min_time_steps' in params else 0
max_time_steps = params[
'max_time_steps'] if 'min_time_steps' in params else 800
integration_step = params['dt']
planner = _deep_smp_module.DSSTMPCWrapper(
system_type=system,
start_state=np.array(data['start_goal'][0]),
# goal_state=np.array(ref_path[-1]),
goal_state=np.array(data['start_goal'][1]),
goal_radius=params['goal_radius'],
random_seed=0,
sst_delta_near=params['sst_delta_near'],
sst_delta_drain=params['sst_delta_drain'],
obs_list=obs_list,
width=params['width'],
verbose=params['verbose'],
mpnet_weight_path=params['mpnet_weight_path'],
cost_predictor_weight_path=params['cost_predictor_weight_path'],
cost_to_go_predictor_weight_path=params[
'cost_to_go_predictor_weight_path'],
num_sample=params['cost_samples'],
shm_max_step=params['shm_max_steps'],
np=params['n_problem'],
ns=params['n_sample'],
nt=params['n_t'],
ne=params['n_elite'],
max_it=params['max_it'],
converge_r=params['converge_r'],
mu_u=params['mu_u'],
std_u=params['sigma_u'],
mu_t=params['mu_t'],
std_t=params['sigma_t'],
t_max=params['t_max'],
step_size=params['step_size'],
integration_step=params['dt'],
device_id=params['device_id'],
refine_lr=params['refine_lr'],
weights_array=params['weights_array'],
obs_voxel_array=obc.reshape(-1))
solution = planner.get_solution()
data_cost = np.sum(data['cost'])
# th = 1.2 * data_cost
# start experiment
tic = time.perf_counter()
for iteration in tqdm(range(number_of_iterations)):
planner.deep_smp_step(
params['refine'],
refine_threshold=params['refine_threshold'],
using_one_step_cost=params['using_one_step_cost'],
cost_reselection=params['cost_reselection'],
goal_bias=params['goal_bias'],
NP=params['n_problem'])
solution = planner.get_solution()
# and np.sum(solution[2]) < th:
if solution is not None or time.perf_counter(
) - tic > params['max_planning_time']:
break
toc = time.perf_counter()
# if solution is not None:
# print(solution[0], solution[2])
# print(mpc_mpnet.costs)
costs = solution[2].sum() if solution is not None else np.inf
result = {
'env_id': env_id,
'traj_id': traj_id,
'planning_time': toc - tic,
'successful': solution is not None,
'costs': costs,
'traj': solution[0]
}
print("\t{}, time: {} seconds, {}(ref:{}) costs".format(
result['successful'], result['planning_time'], result['costs'],
np.sum(data['cost'])))
return result
def experiment(env_id,
traj_id,
verbose=False,
system='cartpole_obs',
params_module=None):
print("env {}, traj {}".format(env_id, traj_id))
obs_list = get_obs(system, env_id)[env_id].reshape(-1, 2)
data = load_data(system, env_id, traj_id)
ref_path = data['path']
start_goal = data['start_goal']
# print(start_goal)
env_vox = np.load('mpnet/sst_envs/{}_env_vox.npy'.format(system))
obc = env_vox[env_id, 0]
# print(obs_list)
params = params_module.get_params()
number_of_iterations = params['number_of_iterations'] #3000000#
min_time_steps = params[
'min_time_steps'] if 'min_time_steps' in params else 80
max_time_steps = params[
'max_time_steps'] if 'min_time_steps' in params else 400
integration_step = params['dt']
planner = _deep_smp_module.DSSTMPCWrapper(
system,
start_state=np.array(ref_path[0]),
goal_state=np.array(ref_path[-1]),
# goal_state=np.array(data['start_goal'][-1]),
goal_radius=params['goal_radius'],
random_seed=0,
sst_delta_near=params['sst_delta_near'],
sst_delta_drain=params['sst_delta_drain'],
goal_bias=params['goal_bias'],
obs_list=obs_list,
width=params['width'],
verbose=params['verbose'],
mpnet_weight_path=params['mpnet_weight_path'],
cost_predictor_weight_path=params['cost_predictor_weight_path'],
cost_to_go_predictor_weight_path=params[
'cost_to_go_predictor_weight_path'],
num_sample=params['cost_samples'],
ns=params['n_sample'],
nt=params['n_t'],
ne=params['n_elite'],
max_it=params['max_it'],
converge_r=params['converge_r'],
mu_u=params['mu_u'],
std_u=params['sigma_u'],
mu_t=params['mu_t'],
std_t=params['sigma_t'],
t_max=params['t_max'],
step_size=params['step_size'],
integration_step=params['dt'],
device_id=params['device_id'],
refine_lr=params['refine_lr'],
)
solution = planner.get_solution()
data_cost = np.sum(data['cost'])
th = 1.2 * data_cost
## start experiment
tic = time.perf_counter()
for iteration in tqdm(range(number_of_iterations)):
# planner.step(min_time_steps, max_time_steps, params['dt'])
if params['hybrid']:
if np.random.rand() < params['hybrid_p']:
planner.step(min_time_steps, max_time_steps, integration_step)
else:
planner.neural_step(
obc.reshape(-1),
params['refine'],
refine_threshold=params['refine_threshold'],
using_one_step_cost=params['using_one_step_cost'],
cost_reselection=params['cost_reselection'])
else:
# planner.step(min_time_steps, max_time_steps, integration_step)
planner.neural_step(
obc.reshape(-1),
params['refine'],
refine_threshold=params['refine_threshold'],
using_one_step_cost=params['using_one_step_cost'],
cost_reselection=params['cost_reselection'])
# planner.mpc_step(integration_step)
solution = planner.get_solution()
if solution is not None: #and np.sum(solution[2]) < th:
break
toc = time.perf_counter()
if solution is not None:
print(solution[0], solution[2])
# print(mpc_mpnet.costs)
costs = solution[2].sum() if solution is not None else np.inf
result = {
'env_id': env_id,
'traj_id': traj_id,
'planning_time': toc - tic,
'successful': solution is not None,
'costs': costs
}
print("\t{}, time: {} seconds, {}(ref:{}) costs".format(
result['successful'], result['planning_time'], result['costs'],
np.sum(data['cost'])))
return result
def experiment(env_id,
traj_id,
verbose=False,
system='cartpole_obs',
params=None):
print("env {}, traj {}".format(env_id, traj_id))
if system in ['cartpole_obs', 'acrobot_obs', 'car_obs']:
obs_list = get_obs(system, env_id)[env_id]
elif system in ['quadrotor_obs']:
obs_list = get_obs_3d(system, env_id)[env_id].reshape(-1, 3)
data = load_data(system, env_id, traj_id)
ref_path = data['path']
ref_cost = data['cost'].sum()
start_goal = data['start_goal']
min_time_steps = params['min_time_steps']
max_time_steps = params['max_time_steps']
integration_step = params['integration_step']
if system == 'quadrotor_obs':
env_constr = standard_cpp_systems.RectangleObs3D
env = env_constr(obs_list, params['width'], 'quadrotor')
else:
print("unkown system")
exit(-1)
planner = _sst_module.SSTWrapper(state_bounds=env.get_state_bounds(),
control_bounds=env.get_control_bounds(),
distance=env.distance_computer(),
start_state=start_goal[0],
goal_state=start_goal[-1],
goal_radius=params['goal_radius'],
random_seed=params['random_seed'],
sst_delta_near=params['sst_delta_near'],
sst_delta_drain=params['sst_delta_drain'])
solution = planner.get_solution()
data_cost = np.sum(data['cost'])
# th = 1.2 * data_cost
## start experiment
tic = time.perf_counter()
for iteration in tqdm(range(params['number_of_iterations'])):
planner.step(env, min_time_steps, max_time_steps, integration_step)
if iteration % 100 == 0:
solution = planner.get_solution()
if (solution is not None and solution[2].sum() <= ref_cost * 1.4
) or time.perf_counter() - tic > params[
'max_planning_time']: #and np.sum(solution[2]) < th:
break
toc = time.perf_counter()
costs = solution[2].sum() if solution is not None else np.inf
result = {
'env_id': env_id,
'traj_id': traj_id,
'planning_time': toc - tic,
'successful': solution is not None,
'costs': costs
}
print("\t{}, time: {} seconds, {}(ref:{}) costs".format(
result['successful'], result['planning_time'], result['costs'],
np.sum(data['cost'])))
return result
def experiment(env_id, traj_id, verbose=False, model='acrobot_obs', params_module=None):
print("env {}, traj {}".format(env_id, traj_id))
obs_list = get_obs(model, env_id)[env_id].reshape(-1, 2)
data = load_data(model, env_id, traj_id)
ref_path = data['path']
start_goal = data['start_goal']
env_vox = np.load('mpnet/sst_envs/data/acrobot_obs_env_vox.npy')
obc = env_vox[env_id, 0]
width = 6
number_of_iterations = 60000
number_of_refine = 120000
min_time_steps, max_time_steps = 1, 200
integration_step = 1e-2
params = params_module.get_params()
planner = _deep_smp_module.DSSTMPCWrapper(
start_state=np.array(ref_path[0]),
goal_state=np.array(ref_path[-1]),
goal_radius=params['goal_radius'],
random_seed=0,
sst_delta_near=params['sst_delta_near'],
sst_delta_drain=params['sst_delta_drain'],
obs_list=obs_list,
width=params['width'],
verbose=params['verbose'],
mpnet_weight_path=params['mpnet_weight_path'],
cost_predictor_weight_path=params['cost_predictor_weight_path'],
cost_to_go_predictor_weight_path=params['cost_to_go_predictor_weight_path'],
num_sample=params['cost_samples'],
ns=params['n_sample'], nt=params['n_t'], ne=params['n_elite'], max_it=params['max_it'],
converge_r=params['converge_r'], mu_u=params['mu_u'], std_u=params['sigma_u'], mu_t=params['mu_t'],
std_t=params['sigma_t'], t_max=params['t_max'], step_size=params['step_size'], integration_step=params['dt'],
device_id=params['device_id'], refine_lr=params['refine_lr']
)
solution = planner.get_solution()
## start experiment
tic = time.perf_counter()
for iteration in tqdm(range(number_of_iterations)):
planner.mpc_step(integration_step)
# planner.step(min_time_steps, max_time_steps, integration_step)
solution = planner.get_solution()
if solution is not None:
break
for iteration in tqdm(range(number_of_refine)):
solution = planner.get_solution()
# if iteration % 1000 == 0:
# print(solution[2].sum(), data['cost'].sum() )
if solution[2].sum() < data['cost'].sum() * 1.2:
break
else:
planner.step(min_time_steps, max_time_steps, integration_step)
toc = time.perf_counter()
# print(mpc_mpnet.costs)
costs = solution[2].sum() if solution is not None else np.inf
result = {
'env_id': env_id,
'traj_id': traj_id,
'planning_time': toc-tic,
'successful': solution is not None,
'costs': costs
}
print("\t{}, time: {} seconds, {}(ref:{}) costs".format(
result['successful'],
result['planning_time'],
result['costs'],
data['cost'].sum()))
return result