Build High Performance Robotic Applications with NVIDIA Isaac ROS Developer Preview 3

Robots in manufacturing facility

Robots are increasing in complexity, with a higher degree of autonomy, a greater number and diversity of sensors, and more sensor fusion-based algorithms. Hardware acceleration is essential to run these increasingly complex workloads, enabling robotics applications that can run larger workloads with more speed and power efficiency.

The mission of NVIDIA Isaac ROS has always been to empower ROS developers with the accelerated computing packages and tools needed to develop high-performance, power-efficient robotics applications. 

NVIDIA is also pioneering accelerated computing into ROS 2, and continuing to deliver improvements with each release. More than 20 hardware-accelerated ROS packages have been added in the last 2 years, with support for the latest ROS 2 distribution.

The team worked with Open Robotics last year to include adaptation and type negotiation to improve the ROS performance on compute platforms that offer hardware accelerators. They also implemented adaptation and type negotiation for Isaac ROS called NITROS (NVIDIA Isaac Transport for ROS)

The latest NVIDIA Isaac ROS Developer Preview 3 (DP3) release offers major updates and enhancements. Read on to learn more.

NVIDIA Isaac ROS Developer Preview 3

NVIDIA Isaac ROS DP3 includes many new features, enabling the ROS community to benefit from hardware acceleration. Highlights include a new map localizer to automatically localize the robot, updated NvBlox with human detection, the new ROS 2 benchmarking tool to realistically benchmark ROS 2 graphs, and open-source NITROS packages. 

Automatically localize in a map in less than a half second

Mobile robots, such as those used in warehouses or service, need to know their initial pose in a map before starting to navigate the space. The most common method to provide this initial pose information to the robot is to manually set it. This method not only adds manual intervention to the whole process, but can also take upwards of 30 seconds. 

With the new Isaac ROS Map Location ROS package, there is no need to manually set the initial position and direction of the robot. This package uses LiDAR range scans to automatically estimate pose in an occupancy grid map in less than a half second. The localizer can work with either planar LiDARs, or 3D LiDAR and can be used to initialize the navigation for mobile robots.

This ROS package is GPU-accelerated and leverages NITROS for creating efficient ROS pipelines with no communication overhead. It is also integrated with Nav2. For more information, see the Isaac ROS Map Localization package on GitHub.

A GIF showing RVIZ on the left, and Isaac Sim scene top view on the right. Initial global pose of the robot is computed in less than 1 second automatically using the occupancy-grid-localizer package, to start navigation. This removed the need to manually provide the initial pose for Nav2.
Figure 1. The NVIDIA Isaac ROS Map Location package automatically calculates initial global pose in less than half a second

Remove people from 3D reconstruction

Isaac ROS NvBlox provides a GPU-accelerated package for 3D reconstruction and a cost map of the environment around the robot using sensor observations. These are useful to path planners for generating collision-free paths. 

Robots navigating among people need to first detect those people and then navigate differently based on proximity to them. Though people should be part of the cost map (to calculate a collision-free path), they should not be part of 3D reconstruction. 3D reconstruction should include only static obstacles. 

The updated Isaac ROS NvBlox package that is part of Isaac ROS DP3 release detects and segments people and provides a person cost map for avoiding collisions with people. It also provides a static cost map for 3D reconstruction in order to avoid collision with static objects. For more details, see the Isaac ROS NvBlox package on GitHub.

Image showing Isaac ROS NvBlox running with RVIZ output of 3D reconstruction with human detection.
Figure 2. Isaac ROS NvBlox package running human detection and removal during 3D reconstruction

Benchmark ROS graphs the right way

Benchmarking ROS-based graphs should reflect the performance under a realistic workload. Benchmarking the entire ROS graph (instead of a particular ROS node) is important, as it will include the message transport costs in RCL indicative of real-world performance. 

Isaac ROS DP3 includes benchmark tooling for ROS 2 in open source. This tooling does not require modification of nodes to measure results, and standardizes input rosbag data sets for independent verification of benchmark results.

Diagram showing ROS 2 benchmarking tool architecture, including data loader node, playback, node, and monitor nodes.
Figure 3. ROS 2 benchmark architecture

The benchmarking tool uses industry best practices and is professionally hardened for throughput and latency measurement of graphs of nodes in real-time robotics applications. Highlights include:

  • Dependable results: Automated performance measurements are performed for multiple seconds N times (default N = 5), discarding minimum and maximum results to reduce variability. Benchmark results are reported in log files for import into your visualization tool of choice.
  • Input data set: Available for download from NGC under CCv4.0 Attribution License, the r2b data set 2023 provides a consistent input to the graph from rosbag. Additional input data can be added when needed.
  • Input image resolutions: With a broad range of computing hardware available, image processing is performed at different resolutions depending on the robotics application.
  • Input and output transport time: Time spent in RCL publishing and receiving messages for inter and intra process is included in the measurement results. This accurately represents what can be expected in a robotics application and avoids inflated results that remove message passing costs.
  • Input and output type adaptation: Input data is injected using standard ROS types, or using type adaptation and type negotiation.
  • Benchmark parameters: The parameters used for testing include data input length, publishing rate, and input size and can be customized with a configuration file.
  • Throughput auto finder: Measuring peak throughput of the graph, with <1% topic drops, requires automatically finding the peak throughput of the graph. The throughout auto finder efficiently finds the input data publishing rate for peak throughput.
  • Real-time latency: Fixed topic publisher rate for real-time measure of latency. This shows what is provided to the real-time system at the target fixed rate whereas throughput shows what peak performance is possible for the robotics application.
  • Cloud native: Measurements can be performed on Kubernetes as part of automated testing or CI/CD nightly testing as part of modern software development. Measurements can also be performed on local developer systems.
  • Opaque testing: Graphs of nodes are tested as binaries with all performance measurement tooling in benchmark directly. This does not modify code in the graph under test and enables performance measurements inclusive of open source to proprietary solutions with the same tools in an unobtrusive way.
  • Transparency: Results in JSON include the parameters used to run the benchmark, including an MD5 of the input rosbag for independent verification of results.

Register for the webinar, Performance Measurement of Robotics Applications with ros2_benchmark to learn more. 

Support for new platforms

Isaac ROS DP3 adds support for the newest members of the NVIDIA Jetson family, NVIDIA Jetson Orin Nano and NVIDIA Jetson Orin NX. It also supports the newly released NVIDIA Jetson Orin Nano Developer Kit. Isaac ROS is now supported on all Jetson Orin and Xavier series of modules and developer kits.

Image showing NVIDIA Jetson AGX Orin Developer Kit, NVIDIA Jetson Orin Nano Developer Kit, NVIDIA Jetson AGX Orin production modules, NVIDIA Jetson Orin NX production modules, and NVIDIA Jetson Orin Nano production modules.
Figure 4. Isaac ROS DP3 supports the Jetson Orin family of modules and developer kits 

The new release also adds support for the recently released NVIDIA Ada Lovelace architecture and NVIDIA GeForce RTX 40 GPUs.

Starting from this release, the majority of the NITROS-accelerated Isaac ROS package is open source. As an ROS developer, you can now extend the packages, fix any issues you encounter, and contribute back to the Isaac ROS community. 


NVIDIA Isaac ROS Developer Preview 3 is a major update, enabling the ROS community to benefit from hardware acceleration to more easily build high-performance, power-efficient robotics applications. Highlights include the new ROS 2 benchmarking tool to realistically benchmark ROS 2 graphs, a new map localizer to automatically localize the robot, open-source NITROS packages, and updated NvBlox with human detection. 

Additional resources

Discuss (0)