Jenga
Many contacts
Lift a loose block and add it to the top of the tower.
steel preset (the default), recordedon an NVIDIA RTX PRO 6000 Blackwell.Run it
cd examples# the default preset (steel), then the othersuv run shelf play jenga# change solver settings liveuv run shelf tune jengaSource: examples/shelf/02_many_contacts/jenga/ · ported from steel-webgpu assets/scenes/jenga (hero_montage export) @ ad03e49
Presets
| Preset | Solver | Status | Checked |
|---|---|---|---|
steeldefault | steel | works | 2026-11-10, newton f6fc82a8b, NVIDIA RTX PRO 6000 Blackwell Server Edition |
Solvers with no row are untested on this example. The full settings are inpresets.toml.
Code
# SPDX-FileCopyrightText: Copyright (c) 2026 Skild AI and the Steel contributors# SPDX-License-Identifier: Apache-2.0# Direct source: steel-webgpu; see presets.toml and THIRD_PARTY.md.# Modified for Steel Shelf: preset-driven execution and shared runner integration.
############################################################################ Example Jenga## The arm picks up a loose block and places it across the top of the tower, continuing the alternating pattern.## Rebuilt from steel-webgpu's exported scene (assets/scenes/jenga: scene.json# and scene.bin, copied next to this file); see common/exported_scene.py. The# visual meshes draw when a steel-webgpu checkout sits next to the shelf.## Command: python -m shelf run jenga############################################################################
import newton.examplesfrom common.exported_scene import ExportedSceneExample
class Example(ExportedSceneExample): file = __file__
if __name__ == "__main__": parser = Example.create_parser() viewer, args = newton.examples.init(parser) newton.examples.run(Example(viewer, args), args)example = "jenga"source = "steel-webgpu assets/scenes/jenga (hero_montage export) @ ad03e49"default = "steel"
[check]# Arguments `python -m shelf check` passes in test mode.args = []
[gallery]short = "Lift a loose block and add it to the top of the tower."title = "Jenga"frame = 420
[xpbd]status = "untested"
[vbd]status = "untested"
[mujoco]status = "untested"
[featherstone]status = "untested"
[semi_implicit]status = "untested"
[feather_pgs]status = "untested"
[steel]status = "works"# 2 substeps x 16 iterations, colored response: 3.8 ms a frame on an RTX PRO 6000 (the port's 4 x 8 with sleeping:# 6.3 ms). The block lands on the tower 2.8-3.8 mm from the top layer's centre, square to it, and the tower shifts# under 1.5 mm, as at 4 x 8 and within 1.9 mm of a 16 x 32 run; penetration 1.0 mm peak, 0.17 mm mean (4 x 8: 0.8 /# 0.13). 2 x 8 and 4 x 4 let the tower creep 3-19 mm, 2 x 12 doubles the mean penetration, point friction# (friction_anchor_beta = 0) slides it 26 mm, and the immediate response costs 2.5x here (500 rows in one world).# Sleeping off: the tower never sleeps under the arm, so it only cost 8%. Joint limits for the fingers, built within# 0.1 of a limit; no joint has a finite velocity limit, so those rows are off.sim = { fps = 60, substeps = 2 }solver = { solve_mode = "matrix_free", iterations = 16, articulated_contact_response = "propagation_colored", enable_joint_limits = true, joint_limit_activation_gap = 0.1, enable_joint_velocity_limits = false, enable_sleeping = false, free_body_constraint_max = 8192, articulation_constraint_max = 1024, propagation_constraint_max = 16384 }checked = "2026-11-10, newton f6fc82a8b, NVIDIA RTX PRO 6000 Blackwell Server Edition"
[kamino]status = "untested"
[implicit_mpm]status = "untested"
[style3d]status = "untested"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.