def generate_initial_condition(config=None, T_aug_enabled=True, N=10):
T_aug = None
if T_aug_enabled:
bound = 0.25
pos_min = np.array([-bound, -bound, 0])
pos_max = np.array([bound, bound, 0])
yaw_min = 0
yaw_max = 2 * np.pi
T_aug = MultiEpisodeDataset.sample_T_aug(
pos_min=pos_min,
pos_max=pos_max,
yaw_min=yaw_min,
yaw_max=yaw_max,
)
q_slider = sample_slider_position(T_aug=T_aug)
d = sample_pusher_position_and_velocity(T_aug=T_aug)
action_sequence = torch.Tensor(d['v_pusher']).unsqueeze(0).expand([N, -1])
return {
'q_slider': q_slider,
'q_pusher': d['q_pusher'],
'v_pusher': d['v_pusher'],
'action_sequence': action_sequence,
}
def generate_initial_condition(
config=None,
T_aug_enabled=True,
push_length=0.4,
N=None,
n_his=0, # whether to add some zero actions at the beginning
randomize_velocity=False,
randomize_sdf=True,
randomize_color=True,
upright=False,
):
T_aug = np.eye(4)
if T_aug_enabled:
bound = 0.25
pos_min = np.array([-bound, -bound, 0])
pos_max = np.array([bound, bound, 0])
yaw_min = 0
yaw_max = 2 * np.pi
T_aug = MultiEpisodeDataset.sample_T_aug(
pos_min=pos_min,
pos_max=pos_max,
yaw_min=yaw_min,
yaw_max=yaw_max,
)
q_slider = sample_object_position(T_aug=T_aug, upright=upright)
vel_avg = 0.2
dt = config['env']['step_dt']
if N is None:
N = math.ceil(push_length / (vel_avg * dt))
d = sample_pusher_position_and_velocity(
T_aug=T_aug, N=N, n_his=n_his, randomize_velocity=randomize_velocity)
action_sequence = None
action_sequence = d['action_sequence']
config_copy = copy.deepcopy(config)
if randomize_sdf:
config_copy['env']['model']['sdf'] = sample_random_mug()
if randomize_color:
config_copy['env']['model']['color'] = sample_random_color()
return {
'q_slider': q_slider,
'q_object': q_slider,
'q_pusher': d['q_pusher'],
'v_pusher': d['v_pusher'],
'action_sequence': action_sequence,
'config': config_copy,
}
def generate_initial_condition(config=None,
T_aug_enabled=True):
T_aug = None
if T_aug_enabled:
aug_config = config['dataset']['data_augmentation']
bound = 0.25
pos_min = np.array([-bound, -bound, 0])
pos_max = np.array([bound, bound, 0])
yaw_min = 0
yaw_max = 2*np.pi
T_aug = MultiEpisodeDataset.sample_T_aug(pos_min=pos_min,
pos_max=pos_max,
yaw_min=yaw_min,
yaw_max=yaw_max,)
q_slider = sample_slider_position(T_aug=T_aug)
d = sample_pusher_position_and_velocity(T_aug=T_aug)
return {'q_slider': q_slider,
'q_pusher': d['q_pusher'],
'v_pusher': d['v_pusher'],
'action_seq': d['action_seq'],
}
def generate_initial_condition(config=None,
T_aug_enabled=True,
N=10):
T_aug = None
if T_aug_enabled:
pos_min = config['eval']['T_aug']['pos_min']
pos_max = config['eval']['T_aug']['pos_max']
yaw_min = config['eval']['T_aug']['yaw_min']
yaw_max = config['eval']['T_aug']['yaw_max']
T_aug = MultiEpisodeDataset.sample_T_aug(pos_min=pos_min,
pos_max=pos_max,
yaw_min=yaw_min,
yaw_max=yaw_max, )
q_slider = sample_slider_position(T_aug=T_aug)
d = sample_pusher_position_and_velocity(T_aug=T_aug)
action_sequence = torch.Tensor(d['v_pusher']).unsqueeze(0).expand([N, -1])
return {'q_slider': q_slider,
'q_pusher': d['q_pusher'],
'v_slider': d['v_pusher'],
'action_sequence': action_sequence,
}