借助 AI 智能体更快地开发轻量级 USD 运行时

import nanousd

RACK_ASSET = "./assets/shelving_unit.usd"   # your part on disk (referenced, not copied)
FORKLIFT_ASSET = "./assets/forklift.usd"

# 1) New stage with warehouse conventions (Z-up, meters)
stage = nanousd.Stage.create()
stage.set_metadata_token("upAxis", "Z")
stage.set_metadata_double("metersPerUnit", 1.0)
stage.set_metadata_token("defaultPrim", "World")
stage.define_prim("/World", "Xform")

# 2) ASSEMBLE: reference an external part into place, once per grid cell.
#    add_reference(asset_path, prim_path="") pulls the part in without copying it.
for row in range(3):
    for col in range(5):
        rack = stage.define_prim(f"/World/Racks/Rack_{row}_{col}", "Xform")
        rack.add_reference(RACK_ASSET)                     # <- assembly via reference
        rack.create_attribute("xformOp:translate", "double3")
        rack.set_vec3d("xformOp:translate", (col * 3.0, row * 6.0, 0.0))
        order.append("xformOp:translate")

# 3) Built-in geometry so the file is non-empty even without external assets:
floor = stage.define_prim("/World/Floor", "Cube")
floor.create_attribute("size", "double")
floor.set_double("size", 1.0)
place(floor, translate=(6.0, 6.0, -0.05), scale=(30.0, 40.0, 0.1))

# 4) MOVE: animate a forklift with a time-sampled translate op (this "moves" geometry).
stage.set_metadata_double("startTimeCode", 0.0)
stage.set_metadata_double("endTimeCode", 96.0)
stage.set_metadata_double("timeCodesPerSecond", 24.0)

forklift = stage.define_prim("/World/Forklift", "Xform")
forklift.add_reference(FORKLIFT_ASSET)
forklift.create_attribute("xformOp:translate", "double3")
waypoints = [(0.0, (0.0, 0.0, 0.0)),
             (48.0, (12.0, 0.0, 0.0)),
             (72.0, (12.0, 18.0, 0.0)),
             (96.0, (0.0, 18.0, 0.0))]
for t, pos in waypoints:
    forklift.set_sample_vec3d("xformOp:translate", t, list(pos))   # keyframe at time t
forklift.create_attribute("xformOpOrder", "token[]")
forklift.set_token_array("xformOpOrder", ["xformOp:translate"])

# 5) Save
stage.write_usda("warehouse.usda")
print("Wrote warehouse.usda")

python -m pip install -e ./nanousd-python

import nanousd

stage = nanousd.Stage.open("scene.usda")

for prim in stage.traverse():
    print(prim.path, prim.type_name, prim.attribute_names())

cube = stage.get_prim_at_path("/World/Cube")
if cube is not None and cube.has_attribute("size"):
    print(cube.read_double("size"))

Author a USD stage for a warehouse AMR: a base transform, a lidar sensor prim, and two wheel meshes brought in as instanceable references to one shared wheel asset. 
Follow the USD Core Spec. 
When you're done, compose it back and show me the resolved prim tree, and flag anything you had to correct to stay spec-compliant.

/World/Lidar           Cube          (sensor placeholder)
/World/WheelL          Xform    [instanceable → /_assets/Wheel]
/World/WheelR          Xform    [instanceable → /_assets/Wheel]

note: your first draft referenced the wheel without marking the
prims instanceable, so each wheel composed as a full unique copy.
Per the Core Spec that defeats scene-graph instancing — set
instanceable = true on both, re-composed, both now share one prototype.

Here's robot.usda straight from our exporter. 
Check whether it's Core-Spec-compliant: compose it, walk the result, and tell me exactly where it diverges from what the spec says should resolve. 
Don't guess from the text — open it and read the composed values.