newton.controllers.ControllerJointImpedance#

class newton.controllers.ControllerJointImpedance(builder, *, default_dof_indices, stiffness, damping, use_gravity_compensation=True, use_coriolis_compensation=True, use_inertia_decoupling=True, has_qdd_feedforward=False, joint_q_idx=None, joint_qd_idx=None, joint_q_des_idx=None, joint_qd_des_idx=None, joint_qdd_idx=None, joint_f_idx=None, device=None, requires_grad=False)[source]#

Bases: ControllerBase

One-step joint-space impedance controller for a batch of robots.

Has an identical input/output interface to ControllerJointImpedanceModelFree — flat 1D sim arrays in, flat 1D torque array out — except that the dynamics terms (mass matrix, gravity force, Coriolis force) are computed internally from the Newton model rather than supplied by the caller.

Supports heterogeneous robot fleets — robots in the batch may have different DOF counts. The builder articulations define the per-robot topology; the controller pads internal buffers to model.max_dofs_per_articulation and skips padding slots in all kernels.

Only 1-DOF joints (Revolute, Prismatic) and zero-DOF Fixed joints are supported. The PD error term q_des - q is only valid for scalar joint coordinates.

Impedance law (terms enabled at construction):

τ = [M(q) if use_inertia_decoupling else I] · (q̈_des + Kp·Δq + Kd·Δq̇)

  • [C(q,q̇)·q̇ if use_coriolis_compensation else 0]

  • [g(q) if use_gravity_compensation else 0]

Parameters:

  • builder (ModelBuilder) – ModelBuilder with N articulations (one per robot). Articulations may have different DOF counts.

  • default_dof_indices (wp.array[wp.uint32]) – Concatenated per-robot index arrays of length sum(dofs per articulation) mapping controller DOF slots to positions in the flat simulation arrays (robot 0’s indices first, then robot 1’s, etc.).

  • stiffness (wp.array2d[wp.float32] | None) – Position-error gain Kp [N/m or N·m/rad], shape (N, max_dofs). Pass a baked array or None to read from inputs.stiffness each step.

  • damping (wp.array2d[wp.float32] | None) – Velocity-error gain Kd [N·s/m or N·m·s/rad]. Same format as stiffness.

  • use_gravity_compensation (bool) – Add gravity generalized forces to τ.

  • use_coriolis_compensation (bool) – Add Coriolis generalized forces to τ.

  • use_inertia_decoupling (bool) – Premultiply the PD term by M(q).

  • has_qdd_feedforward (bool) – Accept a desired-acceleration feedforward via inputs.joint_qdd.

  • joint_q_idx (wp.array[wp.uint32] | None) – Optional index array (same length as default_dof_indices) overriding it for the position read.

  • joint_qd_idx (wp.array[wp.uint32] | None) – Optional index array for velocity read.

  • joint_q_des_idx (wp.array[wp.uint32] | None) – Optional index array for desired position read.

  • joint_qd_des_idx (wp.array[wp.uint32] | None) – Optional index array for desired velocity read.

  • joint_qdd_idx (wp.array[wp.uint32] | None) – Optional index array for feedforward read.

  • joint_f_idx (wp.array[wp.uint32] | None) – Optional index array overriding default_dof_indices for the torque-output write.

  • device (Any) – Warp device.

  • requires_grad (bool) – Whether internal buffers need gradient support.

class Inputs#

Bases: object

Input struct returned by input().

Dynamics fields (mass matrix, gravity, Coriolis) are computed internally and do not appear here.

damping: wp.array2d[wp.float32] | None#

Velocity-error gain Kd [N·s/m or N·m·s/rad], shape (robot_count, max_dofs). None when gains are baked at construction.

joint_q: wp.array[wp.float32]#

Current joint positions [m or rad], flat sim-level array.

joint_q_des: wp.array[wp.float32]#

Desired joint positions [m or rad], flat sim-level array.

joint_qd: wp.array[wp.float32]#

Current joint velocities [m/s or rad/s], flat sim-level array.

joint_qd_des: wp.array[wp.float32]#

Desired joint velocities [m/s or rad/s], flat sim-level array.

joint_qdd: wp.array[wp.float32] | None#

Desired acceleration feedforward [m/s² or rad/s²], flat sim-level array. None unless has_qdd_feedforward=True.

stiffness: wp.array2d[wp.float32] | None#

Position-error gain Kp [N/m or N·m/rad], shape (robot_count, max_dofs). None when gains are baked at construction.

class Outputs#

Bases: object

Output struct returned by output().

joint_f: wp.array[wp.float32]#

Joint torque command [N or N·m], flat sim-level array.

__init__(builder, *, default_dof_indices, stiffness, damping, use_gravity_compensation=True, use_coriolis_compensation=True, use_inertia_decoupling=True, has_qdd_feedforward=False, joint_q_idx=None, joint_qd_idx=None, joint_q_des_idx=None, joint_qd_des_idx=None, joint_qdd_idx=None, joint_f_idx=None, device=None, requires_grad=False)#
input()#

Return a pre-allocated Inputs without dynamics fields.

is_graphable()#
output()#

Return a pre-allocated Outputs with a flat torque array.

step(*, inputs, outputs, dt)#

Run one impedance-control step.

Parameters:

  • inputs (Inputs) – Populated Inputs struct. Dynamics terms are computed internally from the Newton model.

  • outputs (Outputs) – Outputs struct to write torques into.

  • dt (float | wp.array[wp.float32]) – Unused. Accepted for API compatibility.