isaaclab-preparing-assets-for-newton

Prepare Assets For Newton With MJWarp

When To Use

Read the asset migration guide first. This skill follows that page in the same order and targets Newton with MJWarp, not Newton solvers generally. Use the sim-to-sim skill only after the asset and task run in both backends.

1D cable / rod assets are out of scope here: they are authored fresh as Newton deformables (Newton + VBD only), not converted from a PhysX rigid asset. See the Deformables guide.

Implicit MPM particle assets and rigid-MPM coupling are also out of scope; see the Implicit MPM page.

Workflow

1. Multi-backend Asset Importing Pipeline. Use provided Isaac Lab assets directly in PhysX and MJWarp; Newton parses their supported authored USD Physics and PhysX properties. For a new URDF or MJCF, keep run_asset_transformer=True and run_multi_physics_conversion=True so the importer creates neutral, PhysX, and MuJoCo payloads. Account for its nested rigid-body structure.
2. Use per-solver schema fragments. Pass a list of fragments to each spawner field. Put common USD Physics properties in solver-common fragments such as UsdPhysicsRigidBodyCfg. Put MJWarp-specific fields in Mujoco*Cfg, Newton-native fields in Newton*Cfg, and PhysX-only fields in Physx*Cfg. Confirm support in the generated schema APIs.
3. Audit the authored mechanical model. Check every dynamic link and contact-relevant object for intentional mass, COM, inertia and frames, collision geometry, approximation and scale, materials, articulation root, fixed-base and fixed-joint representation, joint axes and limits, self-collision, and gravity overrides.
4. Match contact and friction behavior. Expect more default slip in MJWarp. Validate colliders, material bindings, contact locations and gripper force. Set per-shape condim with MujocoCollisionCfg, tune material friction, then set global MJWarpSolverCfg(cone=..., impratio=...). Treat priority, solmix, solref, and solimp as expert per-collider overrides. Use fixed-grasp displacement, contact count, effort, penetration, and success metrics.
5. Velocity limits distinction. actuator_velocity_limit is a rated speed and joint_velocity_limit is a requested solver clamp. Isaac Lab writes the latter to Newton's model, but MJWarp does not consume or enforce it; Kamino does. Add task or control checks for required speed bounds under MJWarp. Use joint_effort_limit for solver limits and actuator_effort_limit to clip explicit actuator models.
6. Why MJWarp often needs more armature. Use reflected rotor inertia or controlled response tests for articulated coordinates. A plain rigid object has body inertia, not actuator armature. Correct mass, inertia, units, reset penetration, effort, action scale, control period, and contact capacity before changing armature.
7. Retune damping with armature. Increasing armature changes effective inertia and damping ratio. Tune armature, stiffness, and damping together from a step response, use conservative action scales, and keep targets away from hard stops to prevent bang-bang control.
8. Choose an MJWarp starting profile. Do not translate PhysX parameters numerically. Start from the nearest profile on the MJWarp solver page, keep the documented convergence defaults initially, enable NewtonCfg.debug_mode, and use MuJoCo contacts unless the task requires NewtonCfg.collision_cfg. Then run zero and random agents in PhysX and MJWarp through multiple resets. Check non-finite state, first-step impulses, saturation, excessive angular velocity, contact loss, and warnings. Reject penetrations, impossible mimic states, and invalid randomized geometry before stepping.
9. Diagnose MJWarp-only failures. Reproduce the first bad step with one environment, a fixed state, no randomization, and identical actions. Classify initialization/model, contact/capacity, control, or dense-scene failures before tuning. Raise overflowing capacity first; change convergence settings only after the asset, reset, controller, contact model, and capacities are valid.

Validation

Require an asset that parses into both backends with supported per-solver fields, intentional mass, collision and contact data, physically justified actuator parameters, finite task behavior, and valid resets. Do not classify an asset as ready while importer or solver warnings remain unexplained.

Maintenance

Keep this skill synchronized section-for-section with the asset migration guide and use the MJWarp solver page for current profile values.

References

• Compact reference
• Examples
• Evaluations
• MJWarp solver tuning
• Physics schemas: fragments and legacy configuration classes
• Newton Simulation Tuning guide