Constraint-aware routing of electric wires and AC/fluid pipes through a car (USD),
inside NVIDIA Omniverse. An Omniverse Kit extension voxelizes the scene with Warp
and an in-its-own-container cuGraph service solves the routes on the GPU. This is
the v2 redesign that replaces the original cuOpt POC (see omni/cuopt/).
| Piece | Where | What |
|---|---|---|
services/solver/piperouter_solver |
pure Python (GPU-optional) | grids → direction-aware weighted lattice → SSSP → polyline; all constraint math |
services/solver/piperouter_service |
Docker, cuGraph/GPU | FastAPI /solve /solve_all; grids handed over via /dev/shm/piperouter |
exts/omni.piperouter |
Omniverse Kit | omni.ui panel, Warp voxelization + thermal/EM fields, USD tube authoring, the expert workflow |
Constraints: hard = collision + per-wire clearance, thermal melt cutoff, waypoints;
soft (weighted edge cost) = surface-hug, thermal, EM (× wire sensitivity), bend/turn
penalty. Wire types (cost/mass/diameter/bend/temp) live in wire_types.json.
Design + plans: docs/superpowers/specs/2026-06-03-piperouter-cugraph-extension-design.md
and docs/superpowers/plans/2026-06-03-piperouter-m{1,2,3,4}-*.md.
# 1. Start the GPU solver (workstation with an NVIDIA GPU + Container Toolkit)
docker compose up --build -d
curl http://localhost:8000/health # {"status":"ok","backend":"gpu"}
# 2. Enable the extension in USD Composer / Isaac Sim:
# Window > Extensions > add this repo's exts/ to the search paths > enable omni.piperouterSee exts/omni.piperouter/docs/README.md for the full workflow (Route All → refine one
wire with waypoints → lock → BOM export) and docker-compose.yml for the service.
python3 -m venv .venv && .venv/bin/pip install -e services/solver
.venv/bin/pip install numpy scipy pytest fastapi "uvicorn[standard]" httpx usd-core warp-lang
# solver core + service (51 tests)
cd services/solver && ../../.venv/bin/pytest -p no:pqm -q
# extension headless logic (16 tests: Warp voxelize, USD authoring, real-HTTP solve)
cd exts/omni.piperouter && ../../.venv/bin/pytest -p no:pqm -qGPU paths (cuGraph in the container, Warp in the extension) run on the hardware; the solver core falls back to scipy dijkstra so tests pass without a GPU.
M1 (solver core), M2 (GPU service), M3 (extension fields+routing), M4 (refine loop + tagging + overlay + BOM export) are built and tested. Phase 2 (not yet built): cuSolver curvature smoothing, cuOpt multi-wire bundling, hierarchical-corridor lattice, real thermal/EM field solves.