Compatible with the NVIDIA stack.
Ipnops is its own I/O layer for robots, machines, infrastructure and software. It integrates with NVIDIA's simulation tools through OpenUSD, ONNX, MCAP and standard interfaces, as it does with Unreal Engine, ROS 2 and other vendors. Living Worlds Lab, including Ipnops, is a member of the NVIDIA Inception program. NVIDIA names describe what is compatible. They are not endorsements.
What Ipnops brings to the loop.
These exist and work on their own. The tools below plug into them.
Measurement kinds, purposes, system classes and five actions, with four signed documents crossing it: batch, command, receipt and birth record. Live in production.
RunningEvery reading is signed by the device that took it and graded by how that device holds its key: soft, device, se, tee, dice or tpm. A result takes the grade of its weakest input.
SupportedBLE sensor nodes, ESP32-S3 and classic ESP32 microcontrollers, and application-class single-board computers have signed readings into the interface on real hardware.
Verified on hardwareA reading is admitted into a twin only inside the band the twin was solved for. Outside it the reading is recorded and the twin makes no claim.
RunningA system declares what it is for, what it measures and its safe limits. A check returns every issue it finds, each with a code, before anything acts.
RunningSix observe-only bus adapters, for ROS 2, OPC UA, EtherCAT, MQTT, SocketCAN and Modbus, and a sensor-pipeline record sink. Each decodes and signs; none can send.
SupportedObserve-only, shadow, approved. Every command is bounded, expires and is checked against the twin. A person approves what moves a machine, and an advisory stop sits above the machine's own emergency stop.
SupportedSealed data held by a person's passkey or a node's hardware key, and isolated silos where one agent runs with its own kernel and no network.
SupportedAgents, models and jobs run in isolated environments with a signed policy and an identity derived from the node's hardware key, on GPU and CPU machines.
SupportedA REST API, an MCP server and an open SDK client under Apache-2.0.
Supported
From a training task to a twin.
Living Worlds provides the worlds. A task runs in a runtime. Its outputs reach the twin as USD stages and ONNX policies, and the command path to a machine stays behind the gate.
Simulation, end to end.
NVIDIA's simulation tools, item by item, with how each exchanges data with Ipnops. They sit beside Unreal Engine, MuJoCo, Blender and ROS 2, which Ipnops works with in the same way. Living Worlds Lab, including Ipnops, is a member of the NVIDIA Inception program.
Read and write stages with UsdPhysics and semantic labels. Reconstruction writes USDZ beside glTF; a twin's scene moves between tools as USD.
SupportedSimulation-ready OpenUSD assets from NVIDIA's alpha catalogue, imported behind a licence gate. See below.
SupportedGPU simulation kernels, with examples that write USD stages.
SupportedPhysics on GPU and CPU for cloth, rigid and deformable media, with MuJoCo and MuJoCo Warp as solvers.
SupportedTrain, play, rollout, benchmark and export policies, with Newton, Kamino, OV PhysX and PhysX backends. Runs in a rootless container job with GPU access.
SupportedStage rendering.
SupportedWorld models for generation and transfer, behind NVIDIA's guardrail.
SupportedText answers served by the Ipnops models service.
SupportedInference and fine-tuning jobs on LeRobot datasets.
SupportedMotion plans write joint trajectories as MCAP for review before anything is issued.
SupportedPolicy export to ONNX, and LeRobot dataset conversion in both directions.
SupportedSensor replay, and the sensor-pipeline reader that takes operator output from a local socket, observe-only.
SupportedFourier neural operator jobs.
SupportedWeather forecasts.
SupportedRadio maps.
SupportedProximity check.
SupportedRoom impulse responses and a GPU wave-equation engine.
SupportedWorkers advertise the GPUs they hold; jobs name what they need (GPU class, memory, cores, operating system) and place on them.
SupportedA learned-policy runtime contract that never actuates, simulation backend choice, Isaac Lab task descriptors, surrogate-model hosting, typed Cosmos jobs and a detection schema for DeepStream and Isaac ROS pipelines. Built and tested in shadow mode.
SupportedEvaluation and humanoid training frameworks.
Available on requestPost-training and data curation for Cosmos.
Available on requestWhole-body control policies.
Available on requestcuVSLAM, nvblox and FoundationPose beside the ROS 2 adapter.
Available on requestVideo analytics pipelines feeding the interface.
Available on requestOptimised and served inference in runtimes.
Available on request3D deep learning and neural reconstruction.
Available on requestAutonomous-driving simulation and models.
Available on requestRTX camera-sensor rendering.
Available on requestStreamed Kit applications.
Available on requestBoot-measurement and attestation verification is built from vendor documentation. The device run, with a Holoscan Sensor Bridge, is delivered per deployment.
Available on request
SimReady assets, with their licence attached.
Ipnops is compatible with NVIDIA SimReady Central (alpha), one source of simulation-ready OpenUSD assets: browse the catalogue, read each asset's licence, import an asset after you accept it, and get it back checked, with its licence and provenance.
The SimReady Central index of simulation-ready OpenUSD assets, refreshed hourly. Search by keyword and dataset.
SupportedNothing downloads until you name the asset's licence exactly. An unknown or other licence allows evaluation under the maker's terms only.
SupportedEvery file is hash-checked against the maker's bill of materials. Paths, file types, sizes and time are capped. USD is read in a sandboxed process with no network and no program execution, and an import refuses if the sandbox cannot be applied.
SupportedIpnops checks layout, units, default prim, masses and references against the public specification, recomputes mass from the rigid bodies and keeps it beside the maker's declared value.
SupportedThe kept package carries its licence flags, the accepted licence and time, the pinned commit and hashes, and the maker's own validation results.
SupportedNVIDIA's own simready-validate profile results, and placing an imported asset into a twin or world.
Available on requestTasks and models go out to a runtime; USD stages, ONNX policies, trajectories and videos come back as signed outputs.
A job names a kind. No code travels with it: each machine runs only the kinds its owner listed. The job runs in a rootless container with GPU device access and no network.
Outputs are inputs to a twin, not commands. A policy reaches a machine only through the interface: forecast in the twin, purpose check, command gate, approval.
Run it on your GPUs or ours.
Jobs run on the Ipnops fleet or on machines you bring.