unisim.backend.genesis.backend

Host-compatibility implementation of the independent genesis backend.

The adapter serves the SimBackend NumPy contract on top of Genesis 1.3.3, following its measured feasibility mappings (#1372): link-addressed root state (never entity-level getters, REPORT §5.5), control_dofs_position inside an adapter-owned nsteps loop honoring set_pre_step_control (§5.4), host caches refreshed once per step/reset barrier (§5.9), MJCF-named sensor equivalents from link state plus one IMUSensor per accelerometer site with clean (noise-free) data (§3.4/§5.3), genesis-native recoded geom contact masks that must not be compared against MuJoCo tables (§5.10), and a DR capability set restricted to the per-env round-trip-measured items (§3.5 [8] / §5.7).

Classes

GenesisBackend

Independent Genesis backend exposed through the host NumPy profile.
class unisim.backend.genesis.backend.GenesisBackend[source]

Bases: SimBackend

Independent Genesis backend exposed through the host NumPy profile.

Construction performs dependency loading, the MJCF cold-path scan, the process-wide gs.init (once), and scene/entity/sensor creation; materialize() builds the batched solver state and binds all runtime caches. Terrain, geom-name contracts, and site Jacobians fail closed. Native interactive/offscreen rendering attaches lazily post-build (see the play contract section). Call close() to end the process-wide Genesis session; re-initialization afterwards fails closed by design.

Parameters:
__init__(scene, num_envs, sim_dt, *, base_name=None, push_body_name=None, device_id=None, integrator=None, constraint_solver=None, friction_cone=None, solver_iterations=None, **unexpected_kwargs)[source]
Parameters:
materialize()[source]

Build the batched scene, cross-check the import, and bind caches.

Idempotent, and called lazily by the first state access: env constructors that validate state shapes before the explicit lifecycle point (ManagerBasedRlEnv builds its EntityScene before calling materialize(), #1382) work like they do on the MuJoCo backend — the same pattern as the isaacgym backend’s lazy materialize. A closed backend cannot be materialized again.

Return type:

None

get_tensor_execution()[source]

Declare the optional tensor lifecycle without discovering SDKs.

Return type:

TensorExecution

get_tensor_capabilities()[source]

Return fail-closed tensor methods and negotiable state fields.

Return type:

TensorLifecycleCapabilities

get_public_state_widths()[source]

Return canonical qpos/qvel widths for packed tensor reset layout.

Return type:

PublicStateWidths

get_sensor_names()[source]

Return the complete public named-sensor namespace.

Return type:

tuple[str, ...]

get_sensor_inventory()[source]

Return the complete public named-sensor inventory and widths.

Return type:

tuple[SensorDescriptor, ...]

get_state_views(fields=None, device=None)[source]

Return stable qpos/qvel mirrors on Genesis’ exact CUDA device.

Parameters:
Return type:

Mapping[str, Any]

get_sensor_view(name, device=None)[source]

Return one stable supported device-resident sensor view.

Parameters:
Return type:

Any

step_tensor(ctrl, nsteps=1)[source]

Step the single-articulation CUDA profile without a host tensor detour.

Parameters:
Return type:

dict[str, dict[str, float]]

set_state_tensor(env_indices, qpos, qvel, randomization=None)[source]

Commit selected CUDA state while preserving untouched rows.

Parameters:

  • env_indices (Any)

  • qpos (Any)

  • qvel (Any)

  • randomization (TensorResetRandomizationPayload | None)

Return type:

dict[str, dict[str, float]]

set_state_tensor_mask(reset_mask, qpos, qvel, randomization=None, *, full_refresh=False)[source]

Commit full-width CUDA state selected by a persistent bool mask.

Genesis’ public state setters consume a boolean environment mask directly, so the fixed-width reset boundary does not compact selected rows on the host or device.

Parameters:

  • reset_mask (Any)

  • qpos (Any)

  • qvel (Any)

  • randomization (TensorResetRandomizationPayload | None)

  • full_refresh (bool)

Return type:

dict[str, dict[str, float]]

bind_packed_reset(*, device=None)[source]

Begin a device-resident packed reset transaction.

Parameters:

device (Any | None)

Return type:

Any

property num_envs: int

Number of vectorized environments.

property model: Any

Return the backend-owned Genesis rigid entity.

property num_actuators: int

Number of actuators.

property num_dof_vel: int

Number of joint velocity DoFs, excluding the floating base.

get_actuator_ctrl_range()[source]

MJCF ctrlrange metadata; Genesis does not enforce it in-engine.

Return type:

ndarray

get_actuator_names()[source]

Return actuator names in control-vector order on the cold path.

Return type:

tuple[str, ...]

get_actuator_joint_names()[source]

Return each actuator’s target single-DoF joint in control-vector order.

Backends must fail closed when an actuator does not target exactly one hinge/slide joint. Manager action terms use this cold-path metadata to map community joint selectors onto the backend control vector without inspecting backend-private model objects.

Return type:

tuple[str, ...]

get_actuator_gains()[source]

Return per-joint (kp, kd) arrays from the backend model.

Return type:

tuple[ndarray, ndarray]

get_scene_model_file()[source]

Return the materialized scene path for diagnostics, when available.

Return type:

str | None

get_keyframe_qpos(name)[source]

Return the full qpos for a named keyframe, including the floating base.

Parameters:

name (str) – Keyframe name such as "stand" or "home".

Return type:

ndarray

Returns:

Array with shape (nq,).

get_default_qpos()[source]

Return the backend/model default qpos through a stable contract.

Return type:

ndarray

get_default_dof_pos()[source]

Return default joint positions in canonical non-root qpos order.

The returned array is detached, one-dimensional, and excludes floating root coordinates. Backends whose DoF view is actuator-indexed must use that same actuator-target order here.

Return type:

ndarray

get_init_qvel()[source]

Return a zero-initialized qvel vector compatible with set_state_tensor.

Return type:

ndarray

Returns:

Zero-filled qvel array.

get_root_state_layout(root_body_name)[source]

Resolve one body’s floating-root columns on the cold path.

Backends must verify that root_body_name owns a free/floating joint; fixed bodies and runtimes without body-to-root metadata fail closed. Name/model lookup is forbidden on reset and step hot paths, so callers cache either the returned layout or the unsupported result during scene materialization.

Parameters:

root_body_name (str)

Return type:

RootStateLayout

get_body_ids(names)[source]

Resolve body/link names to backend integer IDs.

Parameters:

names (Sequence[str]) – Body/link names.

Return type:

ndarray

Returns:

int32 array with shape (len(names),).

Raises:

ValueError – If any name is not found.

get_motion_body_ids(names)[source]

Resolve body IDs used by motion datasets.

Motion datasets are generated from MuJoCo, so the returned ids follow the MJCF body order with worldbody as id 0, regardless of the backend-native indexing used by get_body_ids.

Parameters:

names (Sequence[str])

Return type:

ndarray

get_geom_size(name)[source]

Return one geom size vector in backend geom-name order.

get_default_geom_sizes returns the complete table of default sizes, while this selector resolves one geom by name.

Parameters:

name (str)

Return type:

ndarray

get_default_geom_sizes()[source]

Return default geometry sizes, shape (ngeom, 3).

Return type:

ndarray

get_geom_contact_masks()[source]

Genesis-native recoded contype/conaffinity of collision geoms.

Genesis re-synthesizes contype/conaffinity at import: the collision matrix semantics are preserved, but the integer values must NOT be compared against MuJoCo tables (REPORT #1372 §5.10).

Return type:

tuple[ndarray, ndarray]

get_geom_friction()[source]

Return audited native Genesis geom-friction coefficients.

Columns follow Genesis’ public coefficient properties as [sliding, torsional, rolling]. The values are captured from exact collision instances during construction; reset mutation and solver feature enablement are separate unsupported semantics.

Return type:

ndarray

get_geom_solref()[source]

Return audited native Genesis contact reference parameters.

Return type:

ndarray

get_geom_solimp()[source]

Return audited native Genesis contact impedance parameters.

Return type:

ndarray

get_geom_id(name)[source]

Resolve one geom name through the backend contract.

Parameters:

name (str)

Return type:

int

get_geom_names()[source]

Return backend geom names in backend id order.

Return type:

tuple[str, ...]

get_geom_body_ids()[source]

Return the owning body id for each geom.

Return type:

ndarray

get_gravity()[source]

Return the backend gravity vector.

Return type:

ndarray

get_body_mass()[source]

Return the backend body-mass table.

Return type:

ndarray

get_body_ipos(env_ids=None)[source]

Return body center-of-mass offsets in each body’s local frame, in meters.

body_ipos[b] is the position of body b’s center of mass in that body’s own local coordinate frame; it is neither an inertia tensor nor a whole-subtree COM.

Without env_ids the canonical model default table is returned with shape (nbody, 3) in backend body-id order. It never changes with reset randomization; fixed-variant backends expose their per-environment default baselines through get_reset_term_default() instead.

With env_ids the current effective per-environment values are returned with shape (len(env_ids), nbody, 3), in env_ids order. The values reflect every reset randomization applied so far, including composition with base_com_offset; an environment untouched by a partial reset keeps its previous values. Backends that do not track per-environment inertial offsets fail closed with NotImplementedError for this form.

Parameters:

env_ids (Sequence[int] | ndarray | None)

Return type:

ndarray

get_dof_damping()[source]

Return cold-captured native Genesis DOF damping in public order.

Return type:

ndarray

get_dof_frictionloss()[source]

Return cold-captured native Genesis friction loss in public order.

Return type:

ndarray

get_dof_armature()[source]

Return the backend dof-armature table.

Return type:

ndarray

get_joint_range(*, names=None)[source]

Return joint position limits, excluding the floating base.

Parameters:

names (Sequence[str] | None) – Optional joint names in requested return order. Adapters that support named scalar joints resolve the model mapping internally.

Return type:

ndarray | None

Returns:

Array with shape (num_dof, 2) and columns [low, high], or None when the backend does not expose limits.

When names is provided by a supporting adapter, the result has shape (len(names), 2). Hinge limits are radians, slide limits are meters, and joints without enabled limits return infinities.

get_joint_dof_indices(names)[source]

Resolve joint names to absolute DoF indices in velocity space.

Parameters:

names (Sequence[str]) – Joint names.

Return type:

ndarray

Returns:

int32 index array with shape (len(names),) addressing the complete qvel vector, including any root coordinates.

get_joint_dof_pos_indices(names)[source]

Resolve joint names to DoF indices in position space (qpos).

Only single-DoF joints are supported; free joints are excluded.

Parameters:

names (Sequence[str]) – Joint names.

Return type:

ndarray

Returns:

int32 index array with shape (len(names),) relative to the joint section of qpos.

get_joint_dof_vel_indices(names)[source]

Resolve joint names to joint-section DoF indices in velocity space.

Unlike get_joint_dof_indices(), the returned indices exclude any root/free-joint velocity columns. They address the same joint section used by get_joint_dof_pos_indices().

Parameters:

names (Sequence[str])

Return type:

ndarray

get_joint_state_qpos_indices(names)[source]

Resolve single-DoF joints to full set_state_tensor qpos columns.

Unlike get_joint_dof_pos_indices(), these indices address the complete qpos vector accepted by set_state_tensor(), including any root coordinates. Manager reset transactions resolve them on the cold path.

Parameters:

names (Sequence[str])

Return type:

ndarray

get_joint_state_qvel_indices(names)[source]

Resolve single-DoF joints to full set_state_tensor qvel columns.

Parameters:

names (Sequence[str])

Return type:

ndarray

get_scene_layout()[source]

Return the immutable materialized public addresses, never native handles.

Return type:

CompiledSceneLayout

get_entity_names()[source]

Return materialized physical entity names in the frozen public order.

Return type:

tuple[str, ...]

get_entity_state(entity)[source]

Return detached root_pose/root_velocity/joint_positions/joint_velocities.

Root fields follow EntityStatePatch’s link-origin/world-frame contract. Non-root joint fields use the frozen entity joint order and native generalized widths. Unavailable fields must fail, never return old data as if current. State freshness follows the adapter’s declared profile.

Parameters:

entity (str)

Return type:

Mapping[str, ndarray]

get_entity_default_state(entity, env_ids=None)[source]

Detached construction/keyframe defaults in selected environment order.

Fields and frames match get_entity_state. Defaults follow immutable variant assignment and never reflect current reset-time randomization. This query does not reset, step or reparse the scene.

Parameters:
Return type:

Mapping[str, ndarray]

reset_entities(request)[source]

Validate the complete selected-entity request, then commit it once.

Validation failure leaves all state unchanged. A native partial commit failure must fault the backend unless the adapter can actually roll back. Unselected entities/environments and fixed identity remain unchanged.

Parameters:

request (Any)

Return type:

None

get_dr_capabilities()[source]

Declare only the per-env round-trip-measured DR items (REPORT §5.7).

Portable mode maps the measured mass, COM, DOF-property and PD-gain terms through audited independent entities. Non-portable mode also declares the solver-level external-force API. Absolute geom_friction randomization remains unsupported because Genesis exposes only a per-env ratio API.

Return type:

DomainRandomizationCapabilities

get_reset_term_default(term)[source]

Return canonical or variant-assigned Genesis reset defaults.

Parameters:

term (str)

Return type:

ndarray

apply_interval_randomization(plan)[source]

Apply a scheduled interval randomization plan.

Generic dispatch: each op yielded by plan.iter_ops() is validated against the builtin term specs (custom terms pass through) and routed to the backend-owned handler table returned by _interval_term_handlers(). Op payloads are torch-first: Torch tensors or TensorIntervalPayload values produced by the caller; host adapters convert tensors at their private boundary through unisim.dr._host, and a missing producer-sampled payload fails closed. A term without a handler fails closed with NotImplementedError naming the backend class and the term. Backends that need per-plan prologue/epilogue semantics (for example clearing staged external forces before the ops accumulate) keep a thin override that calls this base implementation.

Parameters:

plan (IntervalRandomizationPlan)

Return type:

None

apply_body_force(body_ids, force, torque=None)[source]

Apply a world-frame wrench at each body COM for the upcoming step.

Parameters:
Return type:

None

resolve_play_render_plan(*, play_render_mode, play_steps, output_video)[source]

Resolve high-level playback mode into backend-owned render parameters.

Parameters:
Return type:

PlayRenderPlan

bind_renderer(spacing=1.0, *, offset_mode='grid', headless=False, capture=False, width=1280, height=720, camera_kwargs=None)[source]

Lazily attach the Genesis viewer and/or an offscreen camera.

Both are post-build attachments (verified on 1.3.3): the interactive viewer is a genesis.vis.viewer.Viewer built on the visualizer’s shared context; capture uses a visualizer camera built on demand. spacing/offset_mode are accepted for contract parity and ignored: envs are laid out on the Genesis scene’s own grid. The first (headless, capture) pair is pinned, like the isaacgym backend.

Parameters:
Return type:

None

render()[source]

Draw one interactive viewer frame (self-initializes interactive).

Return type:

None

capture_video_frame()[source]

Capture one offscreen RGB frame (self-initializes headless+capture).

Return type:

ndarray

run_playback(*, env, initialize, step, num_steps, output_video=None, render_spacing=None, render_offset_mode=None, headless=None, record_video=None, frame_state_getter=None, camera_kwargs=None, debug_overlay_getter=None, on_frame=None)[source]

Execute backend-owned playback for an env wrapper.

camera_kwargs is normalized into CameraCfg at this boundary; unknown mapping keys fail closed with an error naming them.

debug_overlay_getter is an optional per-frame callback returning a sequence with one entry per environment (len == num_envs); each entry is that env’s sequence of DebugPrimitive (None or empty marks an env without overlay) and returning None disables overlays for the frame. Primitive poses are env-local; the renderer applies grid offsets when composing multiple envs. Backends whose get_play_capabilities().supports_debug_overlay is False fail closed with NotImplementedError when this is not None. On the interactive rendering path only backends whose supports_interactive_debug_overlay is True consume it; the others fail closed with NotImplementedError.

on_frame is an optional per-frame video hook called by offline render pipelines before encoding: it receives (frame_index, frame) with the frame an (H, W, 3) uint8 array, and returns a replacement frame of the same shape/dtype or None to keep the original. Backends rendering through a native (non-offline) renderer fail closed with NotImplementedError when this is not None.

Known boundary: env is the owning env wrapper, not a physics-layer concept. Current playback implementations read env-level configuration (e.g. cfg.scene, cfg.ctrl_dt, cfg.render_spacing) and env-owned playback helpers (get_playback_model, get_physics_state_snapshot) that have no backend-native equivalent yet. The parameter stays on this contract until playback asset/config resolution moves onto backend-owned metadata; backends must only use it on the cold playback path.

Parameters:
Return type:

str | None

get_play_capabilities()[source]

Return backend-native play/render capabilities.

Return type:

PlayCapabilities

get_physics_state_layout()[source]

Return the [time, qpos, qvel] snapshot layout (no mocap tail yet).

Return type:

PhysicsStateLayout

get_physics_state()[source]

Assemble contract [time, qpos, qvel] rows in MJCF order.

The host caches already hold the public layout: whole-MJCF scenes are validated against the source MJCF joint order at materialize time (_validate_non_portable_joint_order()) and portable caches are scattered into the composed-scene column order; Genesis free-base and ball-joint quaternions are natively wxyz, matching MuJoCo. The time column is the backend accumulator advanced by nsteps * sim_dt on every public step path and zeroed on state resets.

Return type:

ndarray

set_physics_state(state)[source]

Restore a get_physics_state snapshot through the set_state path.

Parameters:

state (ndarray)

Return type:

None

get_playback_model(env_index=None)[source]

Return the MJCF file path used by detached MuJoCo playback.

Whole-MJCF scenes replay through their (fragment-expanded) construction source and portable scenes through the composed MJCF. Portable fixed-variant scenes with more than one materialized source follow the Drake/Motrix contract: an explicit env_index is required and the row’s assigned variant scene file is returned.

Parameters:

env_index (int | None)

Return type:

str

close()[source]

End the process-wide Genesis session; re-init afterwards fails closed.

Return type:

None