NVIDIA has released new hardware accelerated packages for the Robot Operating System (ROS). The so-called Isaac GEMs focus on image processing and perception models based on a DNN (Deep Neural Network). They should bring a significant increase in performance compared to pure CPU packages.
NVIDIA Isaac GEMs for ROS
Isaac GEMs are already used in the NVIDIA Isaac robotics development platform. They offer modular options for perception, planning and control that can be integrated into a robotics application. For example, developers can use the perception of obstacles or human speech recognition in robotic applications. Isaac GEMs are included in the Isaac SDK, that for the AI platform NVIDIA Jetson, which is designed for edge computing is optimized.
What is new is the expansion of the range at Isaac GEMs to include the open robot operating system ROS: The new hardware-accelerated packages are intended to reduce the load on the host CPU and offer a clear speed advantage over pure CPU packages. The new Isaac GEMs can be used on NVIDIA Jetson and are available on GitHub. They embrace SGM Stereo Disparity and Point Cloud, Color Space Conversion and Lens Distortion Correction and AprilTags Detection. The first package aims to deal with image pairs from a stereo camera, the second to correct distortions from a monocular camera and the third to detect AprilTags in pictures.
Convergence of NVIDIA and ROS 2
Behind these new opportunities is an agreement between NVIDIA and Open Robotics that aims to improve the performance of ROS 2 on NVIDIA Jetson and GPU-based systems and reduce development time.
The open operating system ROS is now used not only in research, but also in industry, and Open Robotics has been working since last year on the expansion of ROS 2 for interaction with NVIDIA Jetson. In the future, these capabilities will be expanded to enable shared use of memory across GPUs and other processors across NVIDIA Jetson. This should significantly improve the performance of applications that process large amounts of data in real time. NVIDIA cites sensors such as cameras and lidars (light detection and ranging), a radar-like form of optical measurements, as examples.
Other new features announced by NVIDIA include features for ROS developers in Isaac Sim, a scalable robotic simulation tool powered by NVIDIA’s cloud-native Omniverse platform. More GEMs are planned, which NVIDIA intends to present at ROS World on October 21 and 22, 2021.
Details on what’s new for ROS developers can be found on the NVIDIA Developer Blog.