novaphy.solvers.SolverIPC¶

Newton-aligned SolverBase adapter over libuipc for mathematically guaranteed penetration-free Incremental Potential Contact. Requires CUDA ≥ 12.4 and a build with NOVAPHY_WITH_IPC=ON.

When IPC is not built, both SolverIPC and IPCConfig resolve to None; always feature-check before constructing:

if novaphy.has_ipc() and novaphy.solvers.SolverIPC is not None:
    solver = novaphy.solvers.SolverIPC(model, novaphy.solvers.IPCConfig())

How It Works¶

Shape conversion — Boxes, spheres, cylinders, and convex hulls become volumetric tetrahedral meshes; triangle meshes and heightfields become codimensional shells; planes become libuipc ground geometry.
GPU Newton solver — libuipc solves the barrier-based contact problem on GPU.
Barrier contact — a log-barrier potential guarantees no interpenetration at the cost of a fixed dt matching IPCConfig.dt.

The solver advances state through the standard solver.step contract; dt and gravity are validated against the libuipc scene built at construction time, and mismatches raise. The contacts parameter remains present for contract compatibility, but this adapter ignores it because libuipc performs collision internally.

Constructor¶

novaphy.solvers.SolverIPC(model: Model, config: IPCConfig)

Parameter	Description
`model`	Required. `Model` from `ModelBuilder.finalize()`.
`config`	IPCConfig (`dt`, gravity, contact/barrier, material, and Newton-solve parameters).

Build with IPC¶

git submodule update --init --recursive external/libuipc
CMAKE_ARGS="-DNOVAPHY_WITH_IPC=ON -DNOVAPHY_GPU_PLATFORM=NVIDIA -DCMAKE_CUDA_COMPILER=/path/to/nvcc" \
  python -m pip install -e .

See Build from Source for NVIDIA and CoreX toolchain requirements and the standalone CMake preset path.

Example¶

import novaphy

if novaphy.has_ipc():
    config = novaphy.solvers.IPCConfig()
    config.dt = 1.0 / 60.0

    solver   = novaphy.solvers.SolverIPC(model, config)
    state    = model.state()
    control  = model.control()

    for _ in range(steps):
        solver.step(state, state, control, None, config.dt)

When To Use¶

Scenario	Recommendation
Guaranteed no-penetration (precise machinery, jamming, tight clearances)	Use IPC.
Real-time large scenes on commodity GPUs	Often SolverSemiImplicit is faster; profile first.
Deformable bodies (planned)	IPC is the future home; today only rigid.

Limitations¶

Fixed dt requirement: changing dt between steps requires reconstructing the solver.
CUDA-only: no CPU fallback.
There is no public top-level novaphy.IPCWorld container in current releases. New code should use novaphy.solvers.SolverIPC.

Demos¶

Demo	What it shows
`python/demos/demo_ipc_stack.py`	Box stacking with guaranteed no-penetration.