NVIDIA Isaac Sim
NVIDIA Isaac Sim
NVIDIA Isaac Sim, powered by Omniverse, is a scalable robotics simulation application and synthetic data generation tool that powers photorealistic, physically-accurate virtual environments to develop, test, and manage AI-based robots.
Isaac Sim leverages the Omniverse platform’s powerful simulation technologies including advanced GPU-enabled physics simulation with PhysX 5, photorealism with real-time ray and path tracing, and MDL material definition support for physically-based rendering.
Modular Architecture Covers Breadth of Applications
No simulator can address every robotics simulation challenge. But Isaac Sim is built to address many of the most common robotics use cases including manipulation, navigation, synthetic data generation for training data. Furthermore, due to its modular design, the tool can be customized and extended to many new use cases.
Seamless Connectivity and Interoperability
With NVIDIA Omniverse, Isaac Sim benefits from Omniverse Nucleus and Omniverse Connectors, enabling collaborative building, sharing, and importing of environments and robot models in USD.
Easily connect the robot’s brain to a virtual world through Isaac SDK & ROS/ROS2 interface, fully-featured Python scripting, plugins for importing robot and environment models.
Highlighted Feature: Multi-Camera Support for ROS
Generating Synthetic Data for Training Perception Models
Training perception models requires large and diverse datasets. Assembling these datasets can be costly, time-consuming, dangerous and even impossible for certain corner cases. By leveraging the synthetic data generation capabilities of Isaac Sim, developers can bootstrap the training task. In the early phases of a project, synthetic data can accelerate proof of concepts or validate ML workflows. In later stages of a development cycle, real data can be augmented with synthetic data to reduce the time for training a production model. Isaac Sim has built in support for domain randomization allowing for changes in texture, colors, lighting, and placement. It also features support for different types of data including bounding boxes, depth, and segmentation. Developers can output the datasets in KITTI format making it easier to leverage NVIDIA’s Transfer Learning Toolkit (TLT).
One of the key application areas for robots is for manipulators which can identify objects, pick them up and move them. In a modern factory or warehouse environment, manipulators can greatly increase the efficiency and throughput of handling and sorting materials. Issac Sim has built in examples of common tasks like filling bins and stacking bins. These python based examples can be modified to work for your custom tasks. The UR10, Frank Emika Panda (Leonardo), and the low-cost DofBot are manipulator robots supported in Isaac and can jumpstart your projects.
Autonomous mobile robots must be able to move in their environments from point A to point B. This capability is enabled by the navigation stack. Isaac Sim supports the development and testing of your robots navigation capabilities. Isaac Sim provides a complete example of how to use the ROS Navigation stack. In this example, the NVIDIA Carter robot moves its way autonomously through a warehouse environment. Similarly, the Isaac SDK Navigation Stack running on Carter can be easily exercised in one the examples.
Importing Robots/Assets into Isaac Sim
Importing robot models and other assets into robotics simulators is of critical importance and oftentimes a significant challenge when setting up a training or testing scenario. Leveraging the powerful connector capabilities built into Omniverse, Isaac Sim has built-in support for popular product design formats. The advanced URDF importer has been tested on multiple robot models. Additionally, CAD files can be imported directly from Onshape and from STEP files with minimal post-processing.
To make it easier to add assets to different environment, Isaac Sim supports Shapenet. The Shapenet importer provides access to a massive amount of 3D assets.