{"release":{"id":441894382,"tag":"v1.8.0","author":"github-actions[bot]","name":"Release v1.8.0","draft":false,"prerelease":false,"createdAt":"2026-11-11T15:31:26Z","publishedAt":"2026-11-11T15:53:37Z","markdown":"# Newton v1.8.0\r\n\r\nNewton v1.8.0 is a feature release following v1.7.0. It introduces the experimental Steel solver for articulated rigid bodies and adds routed tendons and richer deformable simulation and rendering. If you are upgrading from a previous version, review the upgrade guidance below.\r\n\r\nFor the complete list of changes, see the [changelog](https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#180---2026-11-12).\r\n\r\n## Highlights\r\n\r\n- **New experimental solver: [SolverSteel](https://newton-physics.github.io/newton/latest/api/_generated/newton.solvers.SolverSteel.html).** Developed in collaboration with [Skild AI](https://www.skild.ai/), Steel targets high-throughput, memory-efficient simulation for contact-rich reinforcement learning. Its matrix-free projected Gauss-Seidel solver handles environments containing multiple rigid bodies and articulations, with support for run-to-run determinism, island-based sleeping, and patch-based Coulomb friction. (#4522, #5150)\r\n- **Experimental: [massless routed tendons](https://github.com/newton-physics/newton/blob/release-1.8/docs/tendons.md).** Simulate compliant routes through anchors, pinholes and rollers, including capstan friction and dynamic routing, in XPBD and VBD without discretizing a massive cable. (#4132)\r\n- **[Deformable rest shapes and full-surface contact](https://github.com/newton-physics/newton/blob/release-1.8/docs/concepts/usd_parsing.rst).** Import authored cable, cloth and volume rest shapes independently of their starting poses. Mesh feature contacts and extended VBD rigid-soft Divide-and-Truncate improve handling of coarse deformables against detailed rigid meshes. (#4936)\r\n- **[Filterable Newton diagnostics](https://github.com/newton-physics/newton/blob/release-1.8/docs/guide/diagnostics.rst).** Filter `NewtonWarning` and `NewtonDeprecationWarning` independently of other libraries, while existing built-in warning-category filters continue to work. Library diagnostics also use Python logging so applications can route and capture them. (#4442)\r\n- **Experimental: [skinned deformable visuals](https://github.com/newton-physics/newton/blob/release-1.8/docs/guide/visualization.rst#deformable-visual-meshes).** Attach detailed graphics meshes to cable, cloth or volume simulation without increasing physics resolution. Import USD bind poses and textures, and share evaluated geometry between supported viewers and camera RGB/depth output; use GL or RTX for interactive skinned meshes. (#4968)\r\n- **Experimental: [Kamino solver options](https://github.com/newton-physics/newton/blob/release-1.8/docs/solvers/kamino.rst).** Kamino gains the rigid-body LOX backend and an APGD contact subsolver for DVI. (#4187, #4206)\r\n- **Heterogeneous world layouts.** [ArticulationView](https://newton-physics.github.io/newton/latest/api/_generated/newton.selection.ArticulationView.html) supports batched access to matching articulations across worlds with different contents, including missing matches and uneven array layouts. Set `homogeneous=False` to access selected data that shares a compatible layout. (#4482)\r\n- **[Surface-gripper example](https://github.com/newton-physics/newton/blob/release-1.8/newton/examples/surface_gripper/example_surface_gripper.py).** Build vacuum-style pads with spring-damper seals and load- or separation-based release criteria. Run it with `python -m newton.examples surface_gripper`. (#3974)\r\n\r\n## Performance improvements\r\n\r\nBenchmarks against Newton 1.7.0 were run on Ubuntu 24.04 with CUDA 13.4 and an R615 driver, using each release's Warp version (1.18 and 1.19).\r\n\r\n- **Batched inverse dynamics.** Evaluating mass matrices, gravity and Coriolis terms for 1,024 Franka arms takes 49% less time on GH200 and 62% less on GeForce RTX 5080. (#4704, #5087)\r\n- **VBD cable piles.** Simulating 40 cables for 30 frames takes 26% less time on GH200 and 38% less on GeForce RTX 5080. These results include default-on CUDA fast math; see the numerical-output migration note below. (#4337, #4830)\r\n- **Large-scene construction.** Batched model composition and finalization reduce setup overhead when merging builders or replicating many environments. (#4820)\r\n\r\n## Announcements\r\n\r\n### Upgrade attention\r\n\r\n- **Logging and warnings.** To capture or route library diagnostics, configure handlers on the `newton` logger instead of intercepting `print()` output. Filter Python warnings with `NewtonWarning` and `NewtonDeprecationWarning`; see the [diagnostics guide](https://github.com/newton-physics/newton/blob/release-1.8/docs/guide/diagnostics.rst). (#4442)\r\n- **XPBD stiffness.** If your scene uses multiple solver iterations, springs, bending and tetrahedral constraints can now respond more softly. Increase `ke`, `edge_ke`, or tetrahedral `k_mu`/`k_lambda` as needed and re-test. (#3554)\r\n- **URDF damping in XPBD and VBD.** URDF import now defaults to passive damping, which these solvers do not apply. Pass `joint_damping_mode=\"drive\"` to preserve their imported damping behavior from 1.7. (#4915)\r\n- **Passive springs in Featherstone and SemiImplicit.** Imported MJCF/USD springs now exert forces in these solvers. Recheck assets with authored spring stiffness; set `Model.joint_stiffness` to zero to preserve their previous spring-free behavior. (#4521)\r\n- **Rod layouts and forward kinematics.** Rod state now has seven coordinates and six velocities, and segment frames default to their centers of mass; follow the [Rod migration details](https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#changed) for array indexing, rest poses and local anchors. After a VBD step, use `eval_ik()` to derive current joint state from body poses before calling `eval_fk()`, or exclude those Rod articulations from FK to avoid overwriting their simulated poses. (#3200, #4485)\r\n- **Controller mass matrices.** Mass-matrix evaluation now includes joint armature. Remove code that adds armature to the result again; `include_armature=False` remains a deprecated temporary opt-out. (#4644)\r\n- **Contact friction in Featherstone and SemiImplicit.** Corrected friction limits can reduce sliding friction, and `friction_smoothing` no longer affects the result. Recheck sliding and grasps; tune shape `kf` for low-speed friction and `mu` for the Coulomb limit. (#4913)\r\n- **MuJoCo body requirements.** Jointed bodies now need mass and inertia, either on the body or through fixed attachments; non-root kinematic bodies are rejected. Give moving kinematic roots mass and inertia if their velocity must drive friction or dynamic descendants, since massless roots now use MuJoCo's zero-velocity mocap representation. (#4893)\r\n- **MJCF mesh mass.** Imported meshes now follow MuJoCo's `inertia` mode, including its `legacy` default, so non-convex, open or disconnected meshes can change mass and motion. Recheck those assets; use `inertia=\"exact\"` for signed-volume mass calculation or `\"shell\"` to assign mass to a surface. (#5123)\r\n- **Deformable contact capacity.** If full-surface mesh collision reports overflow, stop and increase `CollisionPipeline.soft_contact_max` to at least the reported requirement before restarting. Missing contacts can defeat VBD's collision protection. (#4936)\r\n- **Experimental: VBD defaults.** The solver now uses compliant ALM for rigid contacts and joints, so recheck scenes tuned with the legacy model. It also uses the USD scene's `newton:maxSolverIterations` when no iteration count is supplied; pass `iterations=10` to keep the previous count. CUDA fast math is enabled by default; pass `enable_fast_math=False` for IEEE division and square root. (#4485, #5027, #4830)\r\n\r\nSee the [full changelog](https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#changed) for specialized migrations, including custom collision buffers, imported welds, MuJoCo actuator properties, MJCF vector defaults and viewer picking.\r\n\r\n### Breaking changes and removals\r\n\r\n- Position targets are always coordinate-sized: remove `use_coord_layout_targets`, index with `joint_q_start`, and use quaternion angular targets for ball/free/distance joints. Velocity targets remain DOF-sized. (#4757)\r\n- Replace removed cable helpers and `add_rod(positions=...)` with `Rod` and `add_rod(rod=...)`; replace `JointType.CABLE`/`add_joint_cable()` with `ROD`/`add_joint_rod()`. Closed chains without explicit rest geometry must repeat their first point at the end. (#4916, #4864)\r\n- Replace the old actuator `Controller*`/`Clamping` names and `controller` fields with their `Drive*`/`ClampingBase` and `drive` replacements. (#5083)\r\n- Remove `legacy_angular_velocity_units` and `legacy_springref_units` from USD calls. Remove compensating application conversions and author angular velocities and spring references in the documented USD units. (#4914)\r\n- MuJoCo CPU requires `use_mujoco_contacts=True`; the formerly accepted Newton-contact combination now raises instead of ignoring supplied contacts. (#4950)\r\n- MJCF ranges under `autolimits=\"false\"` require explicit `limited`/`actuatorfrclimited` flags. Nonstandard body defaults and body `class` attributes are no longer accepted; author body properties directly and use `childclass` for descendants. (#5058, #4691)\r\n\r\n### New deprecations\r\n\r\n- **Torch neural-drive checkpoints.** Migrate `DriveNeuralMLP` and `DriveNeuralLSTM` checkpoints to ONNX with dynamic batch axes and Newton metadata, then remove `allow_unsafe_torch_checkpoints` from USD calls. For ONNX LSTM checkpoints, hidden/cell state uses Warp arrays. (#5090)\r\n- **Joint drive targets.** Replace `target_pos`/`target_vel` keywords and attributes with `target_q`/`target_qd`. (#4521)\r\n- **MuJoCo spring fields.** Use core `joint_stiffness` and coordinate-indexed `joint_rest_q`. The legacy runtime spring-reference alias is available only when its indexing and values match the core array. (#4521, #5133)\r\n- **MuJoCo synchronization interval.** Remove explicit `SolverMuJoCo.update_data_interval` use when adopting per-step input-state synchronization. Explicit intervals remain honored during this deprecation period. (#4984)\r\n- **Style3D configuration.** Pass constructor options through `config=SolverStyle3D.Config(...)` and access `solver.config` fields; `solver.nonlinear_iterations` becomes `solver.config.iterations`. (#4985)\r\n- **Articulation views.** Use `homogeneous=False` instead of `allow_partial_layouts=True`. (#4482)\r\n- **Builder replication.** Replace the `world_count` keyword with `copy_count`; positional calls are unaffected. (#4820)\r\n- **Rod body frames.** Migrate `add_rod(body_frame_origin=\"start\")` to the default `\"com\"` frame; subtract the old `body_com` offset from manually authored body-local anchors. (#4978)\r\n- **Steel solve options.** Use the default `solve_mode=\"serial\"`, or `solve_mode=\"colored\", articulated_contact_response=\"propagation\"` for colored propagation. The old `\"matrix_free\"` and `\"split\"` solve-mode spellings warn and select `\"serial\"`; the dense split algorithm is removed. The `\"propagation_colored\"` contact-response alias warns and selects colored propagation. (#5150)\r\n\r\nThe deprecated mass-matrix and friction options are covered under **Upgrade attention**. See the [complete deprecation list](https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#deprecated) for all affected keywords and attributes.\r\n\r\n### Upcoming removals and platform support\r\n\r\n- **Python 3.10.** Support remains in 1.8 but is deprecated for removal in 1.9; migrate to Python 3.11 or newer. (#4868)\r\n- **Collision-edge aliases.** `Model.mesh_edge_*` and `shape_edge_range` are scheduled for removal in 1.9; use `CollisionPipeline` for model collision detection. (#4770)\r\n- **Windows ARM64.** Experimental support omits optional backends without compatible packages, including MuJoCo and USD. (#4966)\r\n\r\n## Dependency updates\r\n\r\n- The `sim` extra requires MuJoCo and MuJoCo Warp 3.17; USD import requires `newton-usd-schemas>=0.6.0`.\r\n- Release development and validation use Warp 1.19.0; the package minimum remains `warp-lang>=1.17.0`.\r\n- Import/remesh extras require Open3D 0.20 or newer and GLIBC 2.35 or newer on Linux. Open3D-dependent Gaussian PLY import and Poisson remeshing are unavailable on Intel macOS. CoACD requires 1.0.14 or newer; its newer hull-merge behavior can change decompositions.\r\n- RTX requires `ovrtx>=0.5.1.385782` and `ovstage>=0.2.1.385922`. These conflict with Isaac Sim 6.2's standalone development/example pins; use a separate environment or update and validate the constraints together.\r\n- Install the `examples` extra for plotting utilities, which now declare `matplotlib` and `networkx` directly.\r\n\r\nSee the [dependency comparison](https://github.com/newton-physics/newton/compare/v1.7.0...release-1.8?diff=split) for complete requirements.\r\n\r\n## Acknowledgments\r\n\r\nThanks to the Newton maintainers, TSC, project members and community contributors. Special thanks to [Skild AI](https://skild.ai/) and the Newton contributors who collaborated on the Steel solver.\r\n\r\nContributions from outside the Newton maintainer and project-member groups include:\r\n\r\n- @devshahofficial for correcting XPBD stiffness across solver iterations. (#3554)\r\n- @han-xudong for MJCF muscle actuators, compiler inertia guards and URDF mimic-chain import, and @dundysm for camera lighting controls. (#3752, #3742, #4359, #4671)\r\n- @Codnay for preserving particle-body attachment forces in deterministic VBD contact gathering. (#5146)\r\n- @M-Colley for MJCF mesh-inertia modes and component-wise vector-default inheritance. (#5123, #5030)\r\n","html":"<div class=\"markdown-heading\"><h1 class=\"heading-element\">Newton v1.8.0</h1><a id=\"user-content-newton-v180\" class=\"anchor\" aria-label=\"Permalink: Newton v1.8.0\" href=\"#newton-v180\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<p>Newton v1.8.0 is a feature release following v1.7.0. It introduces the experimental Steel solver for articulated rigid bodies and adds routed tendons and richer deformable simulation and rendering. If you are upgrading from a previous version, review the upgrade guidance below.</p>\n<p>For the complete list of changes, see the <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#180---2026-11-12\">changelog</a>.</p>\n<div class=\"markdown-heading\"><h2 class=\"heading-element\">Highlights</h2><a id=\"user-content-highlights\" class=\"anchor\" aria-label=\"Permalink: Highlights\" href=\"#highlights\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<ul>\n<li>\n<strong>New experimental solver: <a href=\"https://newton-physics.github.io/newton/latest/api/_generated/newton.solvers.SolverSteel.html\" rel=\"nofollow\">SolverSteel</a>.</strong> Developed in collaboration with <a href=\"https://www.skild.ai/\" rel=\"nofollow\">Skild AI</a>, Steel targets high-throughput, memory-efficient simulation for contact-rich reinforcement learning. Its matrix-free projected Gauss-Seidel solver handles environments containing multiple rigid bodies and articulations, with support for run-to-run determinism, island-based sleeping, and patch-based Coulomb friction. (#4522, #5150)</li>\n<li>\n<strong>Experimental: <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/docs/tendons.md\">massless routed tendons</a>.</strong> Simulate compliant routes through anchors, pinholes and rollers, including capstan friction and dynamic routing, in XPBD and VBD without discretizing a massive cable. (#4132)</li>\n<li>\n<strong><a href=\"https://github.com/newton-physics/newton/blob/release-1.8/docs/concepts/usd_parsing.rst\">Deformable rest shapes and full-surface contact</a>.</strong> Import authored cable, cloth and volume rest shapes independently of their starting poses. Mesh feature contacts and extended VBD rigid-soft Divide-and-Truncate improve handling of coarse deformables against detailed rigid meshes. (#4936)</li>\n<li>\n<strong><a href=\"https://github.com/newton-physics/newton/blob/release-1.8/docs/guide/diagnostics.rst\">Filterable Newton diagnostics</a>.</strong> Filter <code>NewtonWarning</code> and <code>NewtonDeprecationWarning</code> independently of other libraries, while existing built-in warning-category filters continue to work. Library diagnostics also use Python logging so applications can route and capture them. (#4442)</li>\n<li>\n<strong>Experimental: <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/docs/guide/visualization.rst#deformable-visual-meshes\">skinned deformable visuals</a>.</strong> Attach detailed graphics meshes to cable, cloth or volume simulation without increasing physics resolution. Import USD bind poses and textures, and share evaluated geometry between supported viewers and camera RGB/depth output; use GL or RTX for interactive skinned meshes. (#4968)</li>\n<li>\n<strong>Experimental: <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/docs/solvers/kamino.rst\">Kamino solver options</a>.</strong> Kamino gains the rigid-body LOX backend and an APGD contact subsolver for DVI. (#4187, #4206)</li>\n<li>\n<strong>Heterogeneous world layouts.</strong> <a href=\"https://newton-physics.github.io/newton/latest/api/_generated/newton.selection.ArticulationView.html\" rel=\"nofollow\">ArticulationView</a> supports batched access to matching articulations across worlds with different contents, including missing matches and uneven array layouts. Set <code>homogeneous=False</code> to access selected data that shares a compatible layout. (#4482)</li>\n<li>\n<strong><a href=\"https://github.com/newton-physics/newton/blob/release-1.8/newton/examples/surface_gripper/example_surface_gripper.py\">Surface-gripper example</a>.</strong> Build vacuum-style pads with spring-damper seals and load- or separation-based release criteria. Run it with <code>python -m newton.examples surface_gripper</code>. (#3974)</li>\n</ul>\n<div class=\"markdown-heading\"><h2 class=\"heading-element\">Performance improvements</h2><a id=\"user-content-performance-improvements\" class=\"anchor\" aria-label=\"Permalink: Performance improvements\" href=\"#performance-improvements\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<p>Benchmarks against Newton 1.7.0 were run on Ubuntu 24.04 with CUDA 13.4 and an R615 driver, using each release's Warp version (1.18 and 1.19).</p>\n<ul>\n<li>\n<strong>Batched inverse dynamics.</strong> Evaluating mass matrices, gravity and Coriolis terms for 1,024 Franka arms takes 49% less time on GH200 and 62% less on GeForce RTX 5080. (#4704, #5087)</li>\n<li>\n<strong>VBD cable piles.</strong> Simulating 40 cables for 30 frames takes 26% less time on GH200 and 38% less on GeForce RTX 5080. These results include default-on CUDA fast math; see the numerical-output migration note below. (#4337, #4830)</li>\n<li>\n<strong>Large-scene construction.</strong> Batched model composition and finalization reduce setup overhead when merging builders or replicating many environments. (#4820)</li>\n</ul>\n<div class=\"markdown-heading\"><h2 class=\"heading-element\">Announcements</h2><a id=\"user-content-announcements\" class=\"anchor\" aria-label=\"Permalink: Announcements\" href=\"#announcements\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<div class=\"markdown-heading\"><h3 class=\"heading-element\">Upgrade attention</h3><a id=\"user-content-upgrade-attention\" class=\"anchor\" aria-label=\"Permalink: Upgrade attention\" href=\"#upgrade-attention\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<ul>\n<li>\n<strong>Logging and warnings.</strong> To capture or route library diagnostics, configure handlers on the <code>newton</code> logger instead of intercepting <code>print()</code> output. Filter Python warnings with <code>NewtonWarning</code> and <code>NewtonDeprecationWarning</code>; see the <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/docs/guide/diagnostics.rst\">diagnostics guide</a>. (#4442)</li>\n<li>\n<strong>XPBD stiffness.</strong> If your scene uses multiple solver iterations, springs, bending and tetrahedral constraints can now respond more softly. Increase <code>ke</code>, <code>edge_ke</code>, or tetrahedral <code>k_mu</code>/<code>k_lambda</code> as needed and re-test. (#3554)</li>\n<li>\n<strong>URDF damping in XPBD and VBD.</strong> URDF import now defaults to passive damping, which these solvers do not apply. Pass <code>joint_damping_mode=\"drive\"</code> to preserve their imported damping behavior from 1.7. (#4915)</li>\n<li>\n<strong>Passive springs in Featherstone and SemiImplicit.</strong> Imported MJCF/USD springs now exert forces in these solvers. Recheck assets with authored spring stiffness; set <code>Model.joint_stiffness</code> to zero to preserve their previous spring-free behavior. (#4521)</li>\n<li>\n<strong>Rod layouts and forward kinematics.</strong> Rod state now has seven coordinates and six velocities, and segment frames default to their centers of mass; follow the <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#changed\">Rod migration details</a> for array indexing, rest poses and local anchors. After a VBD step, use <code>eval_ik()</code> to derive current joint state from body poses before calling <code>eval_fk()</code>, or exclude those Rod articulations from FK to avoid overwriting their simulated poses. (#3200, #4485)</li>\n<li>\n<strong>Controller mass matrices.</strong> Mass-matrix evaluation now includes joint armature. Remove code that adds armature to the result again; <code>include_armature=False</code> remains a deprecated temporary opt-out. (#4644)</li>\n<li>\n<strong>Contact friction in Featherstone and SemiImplicit.</strong> Corrected friction limits can reduce sliding friction, and <code>friction_smoothing</code> no longer affects the result. Recheck sliding and grasps; tune shape <code>kf</code> for low-speed friction and <code>mu</code> for the Coulomb limit. (#4913)</li>\n<li>\n<strong>MuJoCo body requirements.</strong> Jointed bodies now need mass and inertia, either on the body or through fixed attachments; non-root kinematic bodies are rejected. Give moving kinematic roots mass and inertia if their velocity must drive friction or dynamic descendants, since massless roots now use MuJoCo's zero-velocity mocap representation. (#4893)</li>\n<li>\n<strong>MJCF mesh mass.</strong> Imported meshes now follow MuJoCo's <code>inertia</code> mode, including its <code>legacy</code> default, so non-convex, open or disconnected meshes can change mass and motion. Recheck those assets; use <code>inertia=\"exact\"</code> for signed-volume mass calculation or <code>\"shell\"</code> to assign mass to a surface. (#5123)</li>\n<li>\n<strong>Deformable contact capacity.</strong> If full-surface mesh collision reports overflow, stop and increase <code>CollisionPipeline.soft_contact_max</code> to at least the reported requirement before restarting. Missing contacts can defeat VBD's collision protection. (#4936)</li>\n<li>\n<strong>Experimental: VBD defaults.</strong> The solver now uses compliant ALM for rigid contacts and joints, so recheck scenes tuned with the legacy model. It also uses the USD scene's <code>newton:maxSolverIterations</code> when no iteration count is supplied; pass <code>iterations=10</code> to keep the previous count. CUDA fast math is enabled by default; pass <code>enable_fast_math=False</code> for IEEE division and square root. (#4485, #5027, #4830)</li>\n</ul>\n<p>See the <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#changed\">full changelog</a> for specialized migrations, including custom collision buffers, imported welds, MuJoCo actuator properties, MJCF vector defaults and viewer picking.</p>\n<div class=\"markdown-heading\"><h3 class=\"heading-element\">Breaking changes and removals</h3><a id=\"user-content-breaking-changes-and-removals\" class=\"anchor\" aria-label=\"Permalink: Breaking changes and removals\" href=\"#breaking-changes-and-removals\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<ul>\n<li>Position targets are always coordinate-sized: remove <code>use_coord_layout_targets</code>, index with <code>joint_q_start</code>, and use quaternion angular targets for ball/free/distance joints. Velocity targets remain DOF-sized. (#4757)</li>\n<li>Replace removed cable helpers and <code>add_rod(positions=...)</code> with <code>Rod</code> and <code>add_rod(rod=...)</code>; replace <code>JointType.CABLE</code>/<code>add_joint_cable()</code> with <code>ROD</code>/<code>add_joint_rod()</code>. Closed chains without explicit rest geometry must repeat their first point at the end. (#4916, #4864)</li>\n<li>Replace the old actuator <code>Controller*</code>/<code>Clamping</code> names and <code>controller</code> fields with their <code>Drive*</code>/<code>ClampingBase</code> and <code>drive</code> replacements. (#5083)</li>\n<li>Remove <code>legacy_angular_velocity_units</code> and <code>legacy_springref_units</code> from USD calls. Remove compensating application conversions and author angular velocities and spring references in the documented USD units. (#4914)</li>\n<li>MuJoCo CPU requires <code>use_mujoco_contacts=True</code>; the formerly accepted Newton-contact combination now raises instead of ignoring supplied contacts. (#4950)</li>\n<li>MJCF ranges under <code>autolimits=\"false\"</code> require explicit <code>limited</code>/<code>actuatorfrclimited</code> flags. Nonstandard body defaults and body <code>class</code> attributes are no longer accepted; author body properties directly and use <code>childclass</code> for descendants. (#5058, #4691)</li>\n</ul>\n<div class=\"markdown-heading\"><h3 class=\"heading-element\">New deprecations</h3><a id=\"user-content-new-deprecations\" class=\"anchor\" aria-label=\"Permalink: New deprecations\" href=\"#new-deprecations\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<ul>\n<li>\n<strong>Torch neural-drive checkpoints.</strong> Migrate <code>DriveNeuralMLP</code> and <code>DriveNeuralLSTM</code> checkpoints to ONNX with dynamic batch axes and Newton metadata, then remove <code>allow_unsafe_torch_checkpoints</code> from USD calls. For ONNX LSTM checkpoints, hidden/cell state uses Warp arrays. (#5090)</li>\n<li>\n<strong>Joint drive targets.</strong> Replace <code>target_pos</code>/<code>target_vel</code> keywords and attributes with <code>target_q</code>/<code>target_qd</code>. (#4521)</li>\n<li>\n<strong>MuJoCo spring fields.</strong> Use core <code>joint_stiffness</code> and coordinate-indexed <code>joint_rest_q</code>. The legacy runtime spring-reference alias is available only when its indexing and values match the core array. (#4521, #5133)</li>\n<li>\n<strong>MuJoCo synchronization interval.</strong> Remove explicit <code>SolverMuJoCo.update_data_interval</code> use when adopting per-step input-state synchronization. Explicit intervals remain honored during this deprecation period. (#4984)</li>\n<li>\n<strong>Style3D configuration.</strong> Pass constructor options through <code>config=SolverStyle3D.Config(...)</code> and access <code>solver.config</code> fields; <code>solver.nonlinear_iterations</code> becomes <code>solver.config.iterations</code>. (#4985)</li>\n<li>\n<strong>Articulation views.</strong> Use <code>homogeneous=False</code> instead of <code>allow_partial_layouts=True</code>. (#4482)</li>\n<li>\n<strong>Builder replication.</strong> Replace the <code>world_count</code> keyword with <code>copy_count</code>; positional calls are unaffected. (#4820)</li>\n<li>\n<strong>Rod body frames.</strong> Migrate <code>add_rod(body_frame_origin=\"start\")</code> to the default <code>\"com\"</code> frame; subtract the old <code>body_com</code> offset from manually authored body-local anchors. (#4978)</li>\n<li>\n<strong>Steel solve options.</strong> Use the default <code>solve_mode=\"serial\"</code>, or <code>solve_mode=\"colored\", articulated_contact_response=\"propagation\"</code> for colored propagation. The old <code>\"matrix_free\"</code> and <code>\"split\"</code> solve-mode spellings warn and select <code>\"serial\"</code>; the dense split algorithm is removed. The <code>\"propagation_colored\"</code> contact-response alias warns and selects colored propagation. (#5150)</li>\n</ul>\n<p>The deprecated mass-matrix and friction options are covered under <strong>Upgrade attention</strong>. See the <a href=\"https://github.com/newton-physics/newton/blob/release-1.8/CHANGELOG.md#deprecated\">complete deprecation list</a> for all affected keywords and attributes.</p>\n<div class=\"markdown-heading\"><h3 class=\"heading-element\">Upcoming removals and platform support</h3><a id=\"user-content-upcoming-removals-and-platform-support\" class=\"anchor\" aria-label=\"Permalink: Upcoming removals and platform support\" href=\"#upcoming-removals-and-platform-support\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<ul>\n<li>\n<strong>Python 3.10.</strong> Support remains in 1.8 but is deprecated for removal in 1.9; migrate to Python 3.11 or newer. (#4868)</li>\n<li>\n<strong>Collision-edge aliases.</strong> <code>Model.mesh_edge_*</code> and <code>shape_edge_range</code> are scheduled for removal in 1.9; use <code>CollisionPipeline</code> for model collision detection. (#4770)</li>\n<li>\n<strong>Windows ARM64.</strong> Experimental support omits optional backends without compatible packages, including MuJoCo and USD. (#4966)</li>\n</ul>\n<div class=\"markdown-heading\"><h2 class=\"heading-element\">Dependency updates</h2><a id=\"user-content-dependency-updates\" class=\"anchor\" aria-label=\"Permalink: Dependency updates\" href=\"#dependency-updates\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<ul>\n<li>The <code>sim</code> extra requires MuJoCo and MuJoCo Warp 3.17; USD import requires <code>newton-usd-schemas&gt;=0.6.0</code>.</li>\n<li>Release development and validation use Warp 1.19.0; the package minimum remains <code>warp-lang&gt;=1.17.0</code>.</li>\n<li>Import/remesh extras require Open3D 0.20 or newer and GLIBC 2.35 or newer on Linux. Open3D-dependent Gaussian PLY import and Poisson remeshing are unavailable on Intel macOS. CoACD requires 1.0.14 or newer; its newer hull-merge behavior can change decompositions.</li>\n<li>RTX requires <code>ovrtx&gt;=0.5.1.385782</code> and <code>ovstage&gt;=0.2.1.385922</code>. These conflict with Isaac Sim 6.2's standalone development/example pins; use a separate environment or update and validate the constraints together.</li>\n<li>Install the <code>examples</code> extra for plotting utilities, which now declare <code>matplotlib</code> and <code>networkx</code> directly.</li>\n</ul>\n<p>See the <a href=\"https://github.com/newton-physics/newton/compare/v1.7.0...release-1.8?diff=split\">dependency comparison</a> for complete requirements.</p>\n<div class=\"markdown-heading\"><h2 class=\"heading-element\">Acknowledgments</h2><a id=\"user-content-acknowledgments\" class=\"anchor\" aria-label=\"Permalink: Acknowledgments\" href=\"#acknowledgments\"><span aria-hidden=\"true\" class=\"octicon octicon-link\"></span></a></div>\n<p>Thanks to the Newton maintainers, TSC, project members and community contributors. Special thanks to <a href=\"https://skild.ai/\" rel=\"nofollow\">Skild AI</a> and the Newton contributors who collaborated on the Steel solver.</p>\n<p>Contributions from outside the Newton maintainer and project-member groups include:</p>\n<ul>\n<li>@devshahofficial for correcting XPBD stiffness across solver iterations. (#3554)</li>\n<li>@han-xudong for MJCF muscle actuators, compiler inertia guards and URDF mimic-chain import, and @dundysm for camera lighting controls. (#3752, #3742, #4359, #4671)</li>\n<li>@Codnay for preserving particle-body attachment forces in deterministic VBD contact gathering. (#5146)</li>\n<li>@M-Colley for MJCF mesh-inertia modes and component-wise vector-default inheritance. (#5123, #5030)</li>\n</ul>\n","assets":[{"contentType":"application/octet-stream","size":8609387,"createdAt":"2026-11-11T15:48:40Z","updatedAt":"2026-11-11T15:48:41Z","downloadCount":5,"downloadUrl":"https://github.com/newton-physics/newton/releases/download/v1.8.0/newton-1.8.0-py3-none-any.whl"}]}}