Randomizing drive gains

73 · Domain randomization

Five arms with a heavy tip on vertical hinges. At 1 s the script writes a sampled joint_target_ke and joint_target_kd per world, calls notify_model_changed(JOINT_DOF_FORCE_PROPERTIES), and steps the target to 90 degrees. Stiffer drives oscillate faster; lower damping overshoots more. Steel integrates drives implicitly by folding dt * kd + dt^2 * ke into the mass matrix, and the notification makes sure that matrix is rebuilt with the new gains.

drag to orbit · scroll to zoom

0.00 s

What to look at

The oscillation period shortens as ke grows.
The arm with the lowest kd overshoots the most.

Key parameter

model.joint_target_ke / joint_target_kd

Per-DOF drive gains. The explicit drive force reads them every step; the implicit part lives in the mass matrix.

Try changing

Use update_mass_matrix_interval=10 and drop the notify call.
Try drive_mode="rows" and compare the responses.

Concept

When Steel sees a randomized value

What it is

Notify with the matching flags after changing these model arrays so Steel refreshes all derived data consistently. Some inputs are cached until notification; mass, inertia and drive gains can also be picked up by periodic mass-matrix rebuilds, so their behavior without notification depends on update_mass_matrix_interval.

In this scene

The arm with the lowest kd overshoots the most.

Considerations

Cost: Call only, default solver, six-link chain and four boxes per world, 1024 worlds (1.12 ms per collide + step): JOINT_DOF_FORCE_PROPERTIES 25 µs (7 µs captured). Flags measured one at a time; combined calls and feature validation can cost more. Body-inertial and joint-DOF notifications also request a mass-matrix refresh.
Graph capture: These flags can be captured in a reset graph. Kinematic-flag changes, sleeping, and shape flags with sparse factors plus restitution run eagerly; captured calls skip some value checks.
Interactions: world_mask limits which worlds lose their warm start; cached data is refreshed for every world.
Use it when: Policies that must tolerate actuator stiffness and damping errors.

The script

cd tutorials && python dr_drive_gains.py prints each variant's final metric; --record out.json also writes the replay. Every tutorial imports replay.py from the same folder.

dr_drive_gains.py
replay.py

# SPDX-FileCopyrightText: Copyright (c) 2026 Skild AI and the Steel contributors
# SPDX-License-Identifier: Apache-2.0

"""Drive gain randomization: per-world joint_target_ke/kd + JOINT_DOF_FORCE_PROPERTIES."""

import math

import newton
import numpy as np
import warp as wp
from newton.solvers import SolverSteel

from replay import Recorder

DT = 1.0 / 240.0
SUBSTEPS = 8
WORLDS = 5


def tf(x, y, z):
    return wp.transform((x, y, z), wp.quat_identity())


def build():
    arm = newton.ModelBuilder()
    link = arm.add_link(xform=tf(0.0, 0.0, 0.6))
    arm.add_shape_box(link, xform=tf(0.15, 0.0, 0.0), hx=0.15, hy=0.02, hz=0.02)
    cfg = newton.ModelBuilder.ShapeConfig(density=8000.0)
    arm.add_shape_sphere(link, xform=tf(0.3, 0.0, 0.0), radius=0.04, cfg=cfg)
    j = arm.add_joint_revolute(
        -1, link, axis=newton.Axis.Z, parent_xform=tf(0.0, 0.0, 0.6), target_ke=20.0, target_kd=2.0
    )
    arm.add_articulation([j])
    builder = newton.ModelBuilder()
    builder.replicate(arm, WORLDS, spacing=(0.0, 0.45, 0.0))
    return builder.finalize()


def simulate(rec, label, ke, kd):
    model = build()
    solver = SolverSteel(model)
    state_0, state_1, control = model.state(), model.state(), model.control()
    newton.eval_fk(model, model.joint_q, model.joint_qd, state_0)

    run = rec.variant(label, model)
    for frame in range(rec.frames(5.0)):
        if frame == rec.fps:  # randomize at 1 s, then step the target to 90 degrees
            model.joint_target_ke.assign(ke.astype(np.float32))  # one DOF per world
            model.joint_target_kd.assign(kd.astype(np.float32))
            # the drives are folded into the mass matrix, which this refreshes
            solver.notify_model_changed(newton.ModelFlags.JOINT_DOF_FORCE_PROPERTIES)
            control.joint_target_q.fill_(math.radians(90.0))
        q = np.degrees(state_0.joint_q.numpy())
        run.frame(state_0, **{f"ke {p:.1f} kd {d:.2f}": v for p, d, v in zip(ke, kd, q)})
        for _ in range(SUBSTEPS):
            solver.step(state_0, state_1, control, None, DT)
            state_0, state_1 = state_1, state_0
    run.check(solver)


if __name__ == "__main__":
    rng = np.random.default_rng(8)
    ke = rng.uniform(2.0, 20.0, WORLDS)
    kd = rng.uniform(0.05, 0.5, WORLDS)
    order = np.argsort(ke)
    ke, kd = ke[order], kd[order]
    rec = Recorder()
    simulate(rec, "gains sampled per world", ke, kd)
    rec.save()

# SPDX-FileCopyrightText: Copyright (c) 2026 Skild AI and the Steel contributors
# SPDX-License-Identifier: Apache-2.0

"""Record Steel tutorial runs as replay JSON for the tutorials page, or print a summary.

Every tutorial imports this module. ``python NN_name.py`` runs the variants and prints
the final metric of each; ``--record out.json`` also writes body poses, shapes and metric
series that the page replays in 3D.
"""

import argparse
import base64
import json
import os
import subprocess
import sys
import time

import newton
import numpy as np
import warp as wp

from source_hash import source_sha256

_RENDERED_TYPES = {
    newton.GeoType.PLANE,
    newton.GeoType.SPHERE,
    newton.GeoType.CAPSULE,
    newton.GeoType.ELLIPSOID,
    newton.GeoType.CYLINDER,
    newton.GeoType.BOX,
    newton.GeoType.MESH,
    newton.GeoType.CONE,
    newton.GeoType.CONVEX_MESH,
}


class Recorder:
    """Collect the variants of one tutorial; ``fps`` is the replay frame rate."""

    def __init__(self, fps: int = 30):
        parser = argparse.ArgumentParser()
        parser.add_argument(
            "--record", metavar="OUT.json", help="write a replay file for the tutorials page"
        )
        parser.add_argument("--device", default=None, help="Warp device, for example cuda:0")
        self.args = parser.parse_args()
        if self.args.device:
            wp.set_device(self.args.device)
        self.fps = fps
        self.script = os.path.abspath(sys.argv[0])
        self.variants: list[Variant] = []

    def frames(self, seconds: float) -> int:
        """Number of replay frames in ``seconds``, including the first."""
        return int(round(seconds * self.fps)) + 1

    def variant(
        self,
        label: str,
        model: newton.Model,
        *,
        ghost: bool = False,
        slot: int | None = None,
        scene: str | None = None,
    ) -> "Variant":
        """Start one run; ``ghost`` overlays run ``slot``, ``scene`` groups runs for charts."""
        variant = Variant(label, model, self.fps, ghost, slot, scene)
        self.variants.append(variant)
        return variant

    def save(self) -> None:
        """Print a summary and, with ``--record``, write the replay file."""
        for variant in self.variants:
            last = variant.metrics[-1] if variant.metrics else {}
            flag = "  (rows dropped)" if variant.overflow else ""
            print(f"{variant.label:<48} {json.dumps(last)}  {variant.wall:.2f} s{flag}")
        if not self.args.record:
            return
        source_sha = source_sha256(self.script)
        doc = {
            "script": os.path.basename(self.script),
            "script_sha256": source_sha,
            "newton_commit": _newton_commit(),
            "newton_version": newton.__version__,
            "warp_version": wp.config.version,
            "gpu": wp.get_device().name,
            "fps": self.fps,
            "variants": [v.to_json() for v in self.variants],
        }
        with open(self.args.record, "w") as f:
            json.dump(doc, f, separators=(",", ":"))
        print(f"wrote {self.args.record}")


class Variant:
    """One recorded run: a pose per frame for every body, plus named metric values."""

    def __init__(self, label, model, fps, ghost, slot, scene):
        self.label, self.model, self.fps, self.ghost, self.slot = label, model, fps, ghost, slot
        self.scene = scene
        self.poses: list[np.ndarray] = []
        self.metrics: list[dict] = []
        self.awake: list[list[int]] = []
        self.particles: list[np.ndarray] = []
        self.overflow = False
        self.start = time.perf_counter()
        self.wall = 0.0

    def frame(self, state: newton.State, awake=None, particles=None, **metrics) -> None:
        """Record ``state.body_q``, optional awake flags, particle positions and metric values."""
        self.poses.append(state.body_q.numpy().astype(np.float32).copy())
        if particles is not None:
            self.particles.append(particles.numpy().astype(np.float32).copy())
        self.metrics.append({k: _round(v) for k, v in metrics.items()})
        if awake is not None:
            self.awake.append([int(x) for x in awake.numpy()])
        self.wall = time.perf_counter() - self.start

    def check(self, solver) -> None:
        """Note whether the solver dropped rows; call once after the run."""
        try:
            solver.check_constraint_capacity()
        except RuntimeError:
            self.overflow = True

    def to_json(self) -> dict:
        poses = (
            np.stack(self.poses)
            if self.poses
            else np.zeros((0, self.model.body_count, 7), np.float32)
        )
        return {
            "label": self.label,
            "fps": self.fps,
            "ghost": self.ghost,
            "slot": self.slot,
            "scene": self.scene,
            "bodies": int(self.model.body_count),
            "frames": int(poses.shape[0]),
            "poses": base64.b64encode(poses.astype(np.float32).tobytes()).decode(),
            "shapes": _shape_table(self.model),
            "metric": self.metrics,
            "awake": self.awake or None,
            "particles": self._particles_json(),
            "wall_s": round(self.wall, 3),
            "overflow": self.overflow,
        }

    def _particles_json(self):
        if not self.particles:
            return None
        frames = np.stack(self.particles)
        tris = self.model.tri_indices
        radius = self.model.particle_radius.numpy()
        return {
            "count": int(frames.shape[1]),
            "frames": base64.b64encode(frames.astype(np.float32).tobytes()).decode(),
            "radius": round(float(radius.mean()), 5) if radius.size else 0.0,
            "tris": [int(v) for v in tris.numpy().reshape(-1)]
            if tris is not None and tris.shape[0]
            else None,
        }


def _shape_table(model):
    types = model.shape_type.numpy()
    scales = model.shape_scale.numpy()
    xforms = model.shape_transform.numpy()
    bodies = model.shape_body.numpy()
    flags = model.shape_flags.numpy()
    shapes = []
    for i, t in enumerate(types):
        if int(t) not in _RENDERED_TYPES or not flags[i] & int(newton.ShapeFlags.VISIBLE):
            continue
        shape = {
            "b": int(bodies[i]),
            "t": int(t),
            "s": [round(float(v), 5) for v in scales[i]],
            "x": [round(float(v), 6) for v in xforms[i]],
        }
        if int(t) in (newton.GeoType.MESH, newton.GeoType.CONVEX_MESH):
            mesh = model.shape_source[i]
            shape["v"] = [round(float(v), 5) for v in np.asarray(mesh.vertices).reshape(-1)]
            shape["i"] = [int(v) for v in np.asarray(mesh.indices).reshape(-1)]
        shapes.append(shape)
    return shapes


def _round(value):
    value = float(value)
    return round(value, 5) if np.isfinite(value) else None


def _newton_commit():
    if os.environ.get("NEWTON_COMMIT"):
        return os.environ["NEWTON_COMMIT"]
    try:
        root = os.path.dirname(os.path.dirname(newton.__file__))
        return subprocess.check_output(["git", "-C", root, "rev-parse", "HEAD"], text=True).strip()
    except (OSError, subprocess.CalledProcessError):
        return None

Recorded with SolverSteel on Newton 1.9.0.dev0, Warp 1.19.0, NVIDIA RTX PRO 6000 Blackwell Server Edition. Replays read the state back every frame, so they are not a speed benchmark. Script · page text

Steel docs and tutorials are licensed under Apache-2.0. Third-party notices. "Skild", "Skild AI" and the Skild logo are trademarks of Skild AI and are not licensed under Apache-2.0.