We are celebrating 2+ years of bringing accelerated computing to ROS and the improvements from feedback requested by companies developing robots for production, researchers, and developers with the release of Isaac ROS 2.0.
Isaac ROS 2.0 is available for ROS2 Humble with major updates to build robotics for commercial deployment, research and development.
This release includes packages for AI perception, image & LIDAR processing, navigation and adds:
- Pre-built Debian packages for ROS 2 Humble on Ubuntu 20.04 for Jetson and other aarch64 platforms from the NVIDIA build farm
- NITROS bridge for optimized ROS 2 ⇔ ROS providing accelerated computing in Noetic (and other ROS versions) with >2x speed up compared to a traditional ROS bridge
- CUDA with NITROS to develop your own CUDA enabled nodes that work with NITROS
- ESS 3.0 model for best in class AI based dense depth perception with confidence mask and a smaller ESS model at >300 fps
- Updated CenterPose NITROS package for object instance pose detection, and corresponding training pipelines in TAO coming in Nov.
- Updated People detection DNN models trained on robot point of view close to the ground
- Added parameters to tune AprilTag detections
- NOVA-Carter reference AMR for ROS 2 Nav2 development and research, see: Nova Carter AMR for ROS 2 w/ 800 megapixel/sec sensor processing for more detail
- new NVIDIA Isaac ROS — isaac_ros_docs documentation site
- Bug fixes and performance improvements
In addition to providing a pre-built Isaac ROS Dev container this release adds pre-compiled aarch64 Ubuntu 20.04 packages for installation on Jetson platforms at isaac.download.nvidia.com/isaac-ros/focal/main.
$ wget -qO - https://isaac.download.nvidia.com/isaac-ros/repos.key | apt-key add -
$ echo ‘deb https://isaac.download.nvidia.com/isaac-ros/focal/main focal main’ | sudo tee /etc/apt/sources.list"
Packages are updated with ROS 2 Humble syncs.
Accelerated computing is essential to move beyond the limits of single threaded performance on a CPU. To take advantage of this, it may not be practical to migrate to a new version of ROS 2 supporting accelerated computing for robotics applications that were developed previously.
The Isaac ROS NITROS bridge brings the value of accelerated computing directly into versions of ROS that don’t have it, with 2x the performance of a traditional ROS bridge.
The newly released ESS 3.0 provides dense AI-based stereo depth and adds a new configurable confidence layer for application tuning. In addition a light ESS model is available to provide depth predictions at 1/16HD with a 3.5x improvement in performance over the 1/4HD model trained from the same dataset.
(ESS 3.0 dense AI-based stereo depth prediction with confidence threshold 0.9 on the r2b_storage dataset)
Isaac ROS 2.0 is available now at github.com/NVIDIA-ISAAC-ROS and is part of our commitment to provide features and hardware acceleration for commercial deployment, development and research for autonomous robots.
Clone the repositories you need into your ROS workspace to build from source with colcon alongside your other ROS2 packages, use pre-built Debian packages, or leverage the pre-built Isaac ROS Dev container. Please note that this release has been tested on the NVIDIA Jetson AGX Orin & Xavier with JetPack 5.1.2.
The next major update will be in March 2024 for NVIDIA’s AI Developer Conference GTC.