When nothing is flat, all patches fall back to the terrain center.

When num_envs < num_cols, should still work without errors.

Fewer than 3 valid points (including zero) falls back to [0, 0, 1].

Test that find_joints_by_actuator_names returns joints in natural joint order. This is a regression test for a bug where joints were returned in ac

Test that find_joints_by_actuator_names returns entity-local indices.

Test find methods with exact and regex matches.

Test find methods with subset filtering.

Removes empty defaults and optionally injects meshdir.

Test fixed-base entity's initial pos/rot are applied to the mocap wrapper.

Test fixed-base mocap entity can have its pose changed at runtime.

Variants must be floating-base; fixed-base raises with a clear message.

Points on z=0 plane → normal = [0, 0, 1].

Flat velocity tasks should have terrain_type='plane' and no terrain_generator.

Foot collision geoms should have specific properties.

Two feet at different heights above a flat platform.

Feet beyond max_distance report max_distance, not -1.

Sensor works correctly across multiple environments.

foot_height observation delegates to sensor.data.heights.

Feet over different terrain heights give different readings.

Foot inside box terrain should report near-zero, not box thickness. Regression: rays inside a box hit the bottom face, producing a bogus height e

Foot below a ground plane should report near-zero, not max_distance.

Test complete scene lifecycle from creation to simulation.

Test that function-based reward terms are not tracked as class terms.

Delayed builtins with same config are fused into one group.

Verify that all G1 actuators have correct control and force limiting. All 29 G1 actuators should have ctrllimited=False (allowing setpoints beyond

G1 velocity env has uniform foot friction across all collision geoms.

G1 tracking tasks should use G1_ACTION_SCALE.

G1 velocity tasks should use G1_ACTION_SCALE.

G1 velocity tasks should have feet/ground and self collision sensors.

Gaussian: mean/std interpretation, values cluster around mean.

TerrainGenerator must not mutate the caller's cfg.num_cols.

Repeated scale operations don't accumulate.

Only specified envs are updated.

dr.geom_size raises ValueError when a non-primitive geom type is selected.

Scaling geom_size updates rbound and aabb consistently.

Events targeting different axes of the same geom compose, not clobber.

Dict- and scalar-shaped curriculum states both yield flat value lists. Regression: the dict branch previously indexed `terms` (a list) by term na

get_base_metadata handles non-actuated joints without KeyError.

_get_export_paths resolves the correct dir, filename, and full path.

Verify correct texture ID detection for textured vs untextured geoms.

Primitive (non-mesh) geoms should always return -1.

get_term_cfg should return resolved config with func as instance.

get_term raises KeyError when the requested term name is not registered.

get_term returns the live term the manager dispatches to, not a copy.

__getitem__ returns lagged frames per-batch (LIFO), clamping when needed.

A box at rest on a plane has its contact normals all vertical.

Test basic Go1 entity properties.

Go1 velocity tasks should use GO1_ACTION_SCALE.

Go1 velocity tasks should have feet/ground and collision sensors.

Finer grid_resolution still produces correct flat patches on a step.

Verify geoms are correctly partitioned by visual compatibility.

All untextured geoms should land in a single group.

Test group history_length overrides term history_length.

height_scan returns correct heights for hits.

height_scan reports max_distance for rays that miss (no ground).

Test that history buffer accumulates observations in correct order.

Verify history captures transient impact forces during a drop. This is the primary use case for the history feature: catching peak forces that oc

Verify history captures forces that match expected physics (F = mg). This test validates that the history buffer stores actual physics values, no

Verify history is None when history_length=0 (default).

Verify index 0 is most recent data in history buffer.

Verify reset clears history for specified environments.

Verify history tensors have correct shape [B, N, H, 3].

Test that clipping is applied before history.

Test that scaling is applied before history.

hold_prob=1.0 keeps lag frozen.

hold_prob=0.0 allows lag to change.

IdealPdActuator computes torques via explicit PD control.

IdealPd clamps computed torques to [-effort_limit, effort_limit].

IdealPd and BuiltinPd produce equivalent actuator forces (qfrc_actuator).

IdealPd and BuiltinPd produce equivalent damping forces with nonzero vel.

IdealPd adds feedforward effort to PD output.

Providing explicit ids populates the corresponding names list.

Verify IK converges kinematically by writing q+dq directly to qpos.

When both names and ids are provided but disagree, a ValueError is raised.

Large ``inductance`` (te >> dt) suppresses early-step torque.

Integrator in position mode ramps torque over time even with ``kp`` and ``kd`` near zero.

In a real env.step(), record_pre_reset fires before _reset_idx zeroes action_manager.action for reset envs. This verifies the hook ordering describe

axes outside valid_axes raises ValueError.

Test accessing non-existent entity raises KeyError.

Verify rotation Jacobian against central finite differences.

Verify position Jacobian against central finite differences.

JointAction with wildcard pattern matches only actuated joints.

IK with joint_limit_weight keeps joints within bounds.

joint_pos_biased property should return joint_pos + encoder_bias.

joint_pos_limits and default_joint_pos_limits should also be inf for unlimited.

Test that joint_pos=None raises error if model has no keyframe.

Test that joint_pos=None works for fixed-base entities with keyframes.

Test that joint_pos=None uses the model's existing keyframe.

joint_pos_rel observation should include encoder_bias.

Test that joint_torques_l2 uses all actuators by default.

Test that joint_torques_l2 only penalizes specified actuators.

joint_vel_rel should NOT include encoder_bias (bias is constant, d/dt = 0).

Joint velocity error is the per-joint RMS of the velocity error.