repository·main·Indexed 25 days ago
https://github.com/isaac-sim/isaacsimA high-fidelity simulation platform built on NVIDIA Omniverse for developing, testing, and training AI-powered robots. It features GPU-accelerated physics, RTX rendering, and workflows for synthetic data generation and reinforcement learning. The documentation covers source installation for Linux and Windows, Python wheel and standalone binary packaging, Docker build environments, and specialized tools like the USD to URDF exporter and the Conveyor Belt Sample using NVIDIA Warp.
The omni.pip.compute extension provides essential general-purpose compute packages required by Isaac Sim extensions. It bundles libraries such as OpenCV (cv2) and other tools used for image processing and numerical computation.
Because it is loaded early in the extension startup sequence, it ensures that these dependencies are available before other extensions attempt to initialize and use them.
isaacsim.ros2.core extension is the foundational C++ backend for ROS 2 integration in Isaac Sim. It provides the Ros2Bridge Carbonite plugin interface, manages per-distribution factory libraries (e.g., Humble, Jazzy), and supplies Python utilities for environment setup, camera information, and namespace collection. All other ROS 2 bridge extensions, such as isaacsim.ros2.bridge, depend on this core extension.MobilityGen is a two-phase pipeline for generating synthetic sensor datasets for mobile robots.
Use this skill when you need robot-trajectory-driven training data. It is distinct from data-collection-sim, which is used for static-scene SDG with randomized object/camera poses but no robot motion.
The isaacsim.replicator.episode_recorder extension is a manifest-first HDF5 recorder and replayer. It allows you to capture simulation state per-episode and replay that state back onto a live USD stage.
It uses a plugin-based architecture via the Recordable protocol. Every recorded channel (such as articulation DOFs, rigid-body velocities, camera intrinsics, or simulation time) is implemented as a Recordable that declares its schema, samples data every tick, and applies it during replay. All metadata is round-tripped through a JSON manifest stored within the HDF5 file.
The isaacsim.core.nodes extension provides OmniGraph nodes for Isaac Sim workflows. It serves as the computational backbone for synthetic data generation and robotics simulation.
Key Capabilities:
Technical Implementation:
isaacsim::core::nodes::CoreNodes) for high performance._isaacsim_core_nodes module.omni.replicator.core and omni.graph.isaacsim.ucx.bridge extension enables high-performance, low-latency communication for distributed simulation scenarios in Isaac Sim. It acts as an integration layer that combines UCX (Unified Communication X) networking infrastructure with node-based processing components to facilitate efficient data transfer between distributed Isaac Sim instances.The Core Experimental API (provided via isaacsim.core.experimental.* extensions) is a rewritten implementation of the standard Isaac Sim Core API. It is designed to be more robust and flexible, moving towards a Warp-based implementation for numeric data.
Key improvements include:
Warp arrays (CPU/GPU) for numeric data containers instead of NumPy or Torch. Methods return Warp arrays and support input from Python basic types, lists, or NumPy arrays.usd, usdrt, fabric, tensor) with a fallback mechanism.Note: This API is experimental and subject to change without deprecation cycles. It is intended to eventually replace the current Core API.
The isaacsim.replicator.nurec_utils module is an experimental toolkit designed for rendering and validating NuRec (neural reconstruction) USD assets within Isaac Sim. It is primarily used for validation, regression testing, and debugging workflows that require rendering NuRec assets from specific camera views and comparing them against reference imagery.
Warning: This module is experimental. APIs, CLI behavior, supported formats, and output file structures are subject to change and are not guaranteed to be future-compatible.
The sample is composed of several specialized modules:
| File | Description |
|---|---|
cb_app.py | Application entry point. Manages the World, simulation loop, and CUDA graph capture. |
cb_scene.py | Builds the conveyor circuit scene and registers velocity fields and rigid bodies. |
cb_scene_building_utils.py | Helpers for USD geometry and rigid body prim configuration. |
cb_actuators.py | Contains VelocityFieldActuator, which runs the Warp kernel for tangential force/torque computation. |
cb_kernels.py | Defines Warp kernels for data preparation, patch assignment, and normal force redistribution. |
cb_conveyor_belt_manager.py | ConveyorBeltManager for registering conveyor belt prims and properties. |
cb_body_manager.py | BodyManager for managing transported rigid bodies and their data. |
cb_material_pair_manager.py | MaterialPairManager for defining friction coefficients via material indices. |
cb_visualizers.py | VelocityFieldVisualizer for debugging belt speeds/directions via debug-draw. |
cb_utils.py | General utility functions. |
The isaacsim.hsb.core extension serves as the backend library for the Holoscan Sensor Bridge (HSB). It provides the core infrastructure for managing sensor data transmission and format conversion.
Key components include:
DLTensor data across two supported data planes: Linux (RoCEv2) or COE (IEEE 1722B).IHsbCore interface, which enables Python bindings for the core functionality.This extension is a required dependency for isaacsim.hsb.nodes and isaacsim.hsb.bridge.
isaacsim.test.collection extension provides a comprehensive integration testing framework for Isaac Sim. It is designed to validate cross-system functionality and core capabilities rather than testing isolated extensions. It ensures that robot simulations, physics engines, rendering components, and various subsystems work correctly in concert.omni.pip.cloud extension provides pre-bundled Python packages optimized for cloud computing environments. It is designed to ensure that essential cloud-related dependencies are available to other extensions automatically, preventing the need for manual pip installations and reducing dependency conflicts during application startup.