Search — docs

MuJoCo's MjSpec.to_xml() can change joint/state-vector ordering after exporting and re-importing modular MJCF using asset/model and attach. This breaks assumptions about stable state ordering across a round-trip.

wp.tile_load_indexed performs an out-of-bounds read when given a negative index. Upper-bound OOB is predicated to zero, but negative indices can dereference before src.data instead of producing 0.

MuJoCo's USD decoder can change inferred mass/inertia even when bodies have explicit inertials and there are visual-only geoms. This breaks MJCF->USD->MuJoCo validation that compares compiled inertials and deterministic rollouts.

Policy logic from an internal Kamino RL drlegs example should be moved into the main DR Legs example with a CLI switch for policy vs standalone. The request prefers moving to ONNX to avoid a torch dependency and to align the benchmark with the example.

Isaac Lab's articulation ordering relies on symbolic convention inference at env creation, which is fragile to backend default changes and adds ~9.3% startup cost. The issue proposes storing ordering metadata in checkpoints and adding a from_checkpoint resolution mode.

On Windows with Isaac Sim 6.0.1 prebuilt binaries, the VS Code setup task generates python.analysis.extraPaths that conflicts with pyproject.toml in a newly created Isaac Lab project. This produces a VS Code/Pylance configuration conflict rather than a clean setup.

Users request a design document covering API Capture, serialized .wrp graphs, and CPU capture/replay. The feature set has expanded enough that code and user guide usage examples are not sufficient to understand behavior and constraints.

MuJoCo’s USD decoder misses physics-purpose material bindings on colliders when different materials are used for visual vs physics purposes. This breaks contact property preservation in MJCF↔USD roundtrip validation.

MuJoCo’s USD decoder does not preserve several semantics needed for roundtrip validation: sites, model names, unlimited joints, and disabled-collider metadata. This undermines compiled-model comparisons and deterministic rollouts after import.

User hits ERROR_OUT_OF_DEVICE_MEMORY on a GCP setup using an NVIDIA L4 while following the recommended Isaac Sim cloud setup guide. The issue occurs with both compiled-from-source and precompiled binaries, preventing progress.

Mesh SDF generation settings are partially configurable and inconsistent across USD/URDF/MJCF imports. Request is to expose configuration paths for Mesh.build_sdf() settings and avoid import paths that drop sdf_* options before expensive cooks occur.

Automatic USD mesh approximation calls ModelBuilder.approximate_meshes() without forwarding method-specific settings. This blocks users from applying required settings (e.g., CoACD threshold migration) through ModelBuilder.add_usd() while respecting USD-authored approximation routing.

wp.quat_twist_angle() uses acos on the twist scalar component, which rounds to 1.0 in float32 near zero, producing a dead zone and discontinuity. Request is to stabilize behavior near zero and add a signed variant.

A user reports the vbd_rigid_rigid_contact example is hard to understand and questions its value, noting it looks like simple wireframe shapes with unclear purpose. They ask for clarification on what is being shown and why.

Warp array annotation repr() renders as a constructor-like string rather than subscript syntax, so it doesn't round-trip with eval(). This affects Sphinx autodoc return type rendering.

Request to warn when CollisionPipeline auto-sizes a very large rigid contact buffer allocation. The warning should include the resolved slot count/bytes, drivers (pair count, world count), and remediation (set rigid_contact_max explicitly).

SolverXPBD lacks docstrings for constructor parameters and several user-facing methods. The issue requests comprehensive Google-style Args documentation and method docstrings.

In Isaac Lab camera output example, point cloud generation fails because camera pose fields become zero/NaN. The report uses Isaac Lab main with Isaac Sim 5.1 and the documented camera output tutorial script.

MuJoCo's `mj_geomDistance` returns exactly 0.0 for separated convex mesh pairs under nativeccd, and the libccd fallback also violates a separating-plane lower bound. The issue is reproduced across multiple MuJoCo versions and persists on current main per the report.

Isaac Lab pip installation docs recommend torch 2.10 while Isaac Sim 6.0 uses torch 2.11 per its python_packages.toml. The user is asking whether the docs should move from 2.10 to 2.11, signaling potential version mismatch confusion.

Documentation/support is inconsistent for free joints applied to non-root links: URDF loader allows it, MJCF does not, USD loader cannot produce it, while the Newton API supports it. The issue asks to clarify intended support, align loaders, and add tests/warnings.

The documentation/type annotation for the mask parameter in eval_jacobian and eval_mass_matrix is unclear (bool array vs array2d[bool], similar to eval_fk). The issue requests updating the annotation and clarifying supported mask formats.

Request to add a site-attached lidar sensor that casts a configurable spherical fan of rays against the model BVH and returns per-ray hit distances (or -1.0 on miss). Proposed API includes azimuth/elevation configuration and min/max range handling.

Proposal to add an Allegro hand in-hand cylinder rotation RL task with grasp-cache initialization, tuned reset poses, contact-aware rewards, a gravity curriculum, and RSL-RL train/play support. The goal is coordinated multi-finger rotation rather than relying on a single pinch contact.

MJX-Warp user hits a JAX UnexpectedTracerError when using a jnp.where-based autoreset that mixes reset data with stepped Warp data, possibly interacting with vmap. They provide a reproducer and are unsure whether this is a bug or interface misuse.

World models are increasingly used in embodied intelligence and generative simulation, yet their scope remains ambiguous across communities. This tutorial presents a design-space view of world models as action-conditioned predictive models that estimate the future evolution of task-relevant observations or states. We categorize existing methods into observation-space and state-space world models, comparing their trade-offs in visual fidelity, spatial structure, physical interpretability, and control usability. We further introduce world action models, which connect predicted futures with executable robot actions, and summarize four representative paradigms: imagine-then-execute, video-feature-conditioned action prediction, joint video-action modeling, and auxiliary video prediction for policy learning. The goal of this tutorial is to clarify the conceptual scope of world (action) models and provide a structured taxonomy for embodied prediction and control.