newton.ik.IKObjective#

class newton.ik.IKObjective[source]#

Bases: object

Abstract base class for inverse kinematics objectives.

Subclasses must implement the following methods:

residual_dim(): Returns the number of residuals (constraints) this objective contributes.
compute_residuals(body_q, joint_q, model, residuals, start_idx): Computes the residuals for this objective.
compute_jacobian_autodiff(tape, model, jacobian, start_idx, dq_dof): Computes the Jacobian using autodiff.

Optional methods for analytic Jacobian support:

supports_analytic(): Returns True if analytic Jacobian is supported, otherwise False.
compute_jacobian_analytic(body_q, joint_q, model, jacobian, joint_S_s, start_idx): Computes the analytic Jacobian if supported.

Device and buffer management:

bind_device(device): Binds the objective to a specific device.
init_buffers(model, jacobian_mode): Initializes any buffers required for Jacobian computation.

Notes

The interface is designed for batch processing of multiple IK problems in parallel.
Each subclass may store per-problem data (e.g., targets) as device arrays.

bind_device(device)#

Binds the objective to a specific device (e.g., CUDA or CPU).

compute_jacobian_analytic(body_q, joint_q, model, jacobian, joint_S_s, start_idx)#

Computes the analytic Jacobian for this objective, if supported.

Parameters:

body_q (wp.array2d(dtype=wp.transform)) – Array of body transforms for each problem.
joint_q (wp.array2d(dtype=wp.float)) – Array of joint coordinates for each problem.
model (newton.Model) – The kinematic model.
jacobian (wp.array3d(dtype=wp.float)) – Output array for the Jacobian.
joint_S_s (wp.array2d(dtype=wp.spatial_vector)) – Motion subspace matrices.
start_idx (int) – Starting index in the Jacobian for this objective.

compute_jacobian_autodiff(tape, model, jacobian, start_idx, dq_dof)#

Computes the Jacobian for this objective using automatic differentiation.

Parameters:

tape (wp.Tape) – Autodiff tape.
model (newton.Model) – The kinematic model.
jacobian (wp.array3d(dtype=wp.float)) – Output array for the Jacobian.
start_idx (int) – Starting index in the Jacobian for this objective.
dq_dof (wp.array2d(dtype=wp.float)) – Derivative of q with respect to DoF.

compute_residuals(body_q, joint_q, model, residuals, start_idx)#

Computes the residuals for this objective and writes them into the residuals array.

Parameters:

body_q (wp.array2d(dtype=wp.transform)) – Array of body transforms for each problem.
joint_q (wp.array2d(dtype=wp.float)) – Array of joint coordinates for each problem.
model (newton.Model) – The kinematic model.
residuals (wp.array2d(dtype=wp.float)) – Output array for residuals.
start_idx (int) – Starting index in the residuals array for this objective.

init_buffers(model, jacobian_mode)#

Initializes any buffers required for Jacobian computation.

Parameters:

model (newton.Model) – The kinematic model.
jacobian_mode (IKJacobianMode) – The Jacobian computation mode (analytic or autodiff).

residual_dim()#: Returns the number of residuals (constraints) this objective contributes. Must be implemented by subclasses.

supports_analytic()#: Returns True if this objective supports analytic Jacobian computation. Subclasses should override if analytic Jacobian is available.