curobo.opt.particle.particle_opt_utils module

class SquashType(value)

Bases: Enum

An enumeration.

CLAMP = 0

CLAMP_RESCALE = 1

TANH = 2

IDENTITY = 3

scale_ctrl(

ctrl,

action_lows,

action_highs,

squash_fn: SquashType = SquashType.CLAMP,

)

get_stomp_cov(

horizon: int,

d_action: int,

tensor_args=TensorDeviceType(device=device(type='cuda', index=0), dtype=torch.float32, collision_geometry_dtype=torch.float32, collision_gradient_dtype=torch.float32, collision_distance_dtype=torch.float32),

cov_mode='acc',

RETURN_M=False,

)

Computes the covariance matrix following STOMP motion planner

Coefficients from here: https://en.wikipedia.org/wiki/Finite_difference_coefficient More info here: https://github.com/ros-industrial/stomp_ros/blob/7fe40fbe6ad446459d8d4889916c64e276dbf882/stomp_core/src/utils.cpp#L36

get_stomp_cov_jit(

horizon: int,

d_action: int,

cov_mode: str = 'acc',

device: device = device(type='cuda', index=0),

)

gaussian_logprob(

mean,

cov,

x,

cov_type='full',

)

Calculate gaussian log prob for given input batch x :param mean (np.ndarray): :type mean (np.ndarray): [N x num_samples] batch of means :param cov (np.ndarray): :type cov (np.ndarray): [N x N] covariance matrix :param x (np.ndarray): :type x (np.ndarray): [N x num_samples] batch of sample values

Returns:

log_prob (np.ndarray)

Return type:

[num_sampls] log probability of each sample

gaussian_logprobgrad(

mean,

cov,

x,

cov_type='full',

)

gaussian_entropy(cov=None, L=None)

Entropy of multivariate gaussian given either covariance or cholesky decomposition of covariance

gaussian_kl(

mean0,

cov0,

mean1,

cov1,

cov_type='full',

)

KL-divergence between Gaussians given mean and covariance KL(p||q) = E_{p}[log(p) - log(q)]

cost_to_go(

cost_seq: Tensor,

gamma_seq: Tensor,

only_first: bool = False,

) → Tensor

Calculate (discounted) cost to go for given cost sequence

cost_to_go_np(cost_seq, gamma_seq)

Calculate (discounted) cost to go for given cost sequence

matrix_cholesky(A)

batch_cholesky(A)