NVIDIA Isaac Manipulator
NVIDIA Isaac⢠Manipulator, built on Isaac ROS, is a reference workflow of NVIDIA-accelerated libraries and AI models. It's designed to help robotics software developers build AI-enabled robot armsâor manipulatorsâthat can perceive, understand, and interact with their environments. This gives you a powerful tool for accelerating solutions for dynamic challenges such as machine tending, high-mix bin picking, inspection, and assembly tasks.
You can integrate any of the Isaac Manipulator libraries or foundation models into your software stacks, bringing accelerated performance and accuracy to your own platforms.
How Isaac Manipulator Works
Isaac Manipulator is comprised of NVIDIA® CUDA®-accelerated libraries, AI models, and reference workflows for robot developers. The software uses cuMotion, a motion planner running multiple trajectory optimizations simultaneously to return the best solution. Isaac Manipulator also uses state-of-the-art foundation models for 6D pose estimation and tracking of novel objects, as well as object detection in indoor environments.
Key Features
Isaac Manipulator brings new levels of dexterity and modular AI capabilities to robotic arms that face limitations in handling intricate tasks and dynamic environments due to their limited adaptability.
Highly Accurate and Performant Modules
Access a collection of advanced, modular packages that enable frictionless kinematics and AI perception for industrial robotic arms.
Accelerated Robot Motion Generation
NVIDIA cuMotion with a MoveIt extension lets you implement custom algorithmic modules without a large pipeline overhead. cuMotion is built with the help of NVIDIA cuRobo.
Faster Development Time
Speed up robotic task implementations with pretrained foundational models that can estimate and track poses of objects, predict ideal grasp points, run robotic arm trajectory optimizations, and more.
cuMotion
NVIDIA cuMotion is a CUDA-accelerated library that helps solve robot motion planning problems at scale by running multiple trajectory optimizations simultaneously to return the best solution.
FoundationPose
This is a state-of-the-art foundation model for 6D pose estimation and tracking of novel objects. It gives you a new method for estimating and tracking the pose of unseen objects and is robust enough for challenging object properties (textureless, glossy, tiny), as well as scenes with fast motion or severe occlusions.
FoundationGrasp
Trained entirely on synthetic data, FoundationGrasp is a performant transformer model that makes dense grasp predictions given an unknown 3D object asset. These grasps can then be executed on the robot with a motion generator such as cuMotion. The model currently supports suction and parallel-jaw grasps.
Coming Soon
SyntheticaDETR
This pre-trained model for object detection in indoor environments allows faster detection, rendering, and training of new objects. It can also be used as a front end to pose estimators like FoundationPose, so it can localize objects using 2D bounding boxes before pose estimation.
Open Ecosystem
Built on ROS
NVIDIA Isaac Manipulator is built on the open-source ROS 2 (Robot Operating System) software framework. This lets the millions of developers in the ROS community easily take advantage of NVIDIA-accelerated libraries and AI models to accelerate their AI robot development and deployment workflows.
Learn More About Isaac ROSIsaac Manipulator Learning Library
More Resources
Ecosystem
Our industry partners are integrating NVIDIA Isaac Manipulator and accelerated computing into their platforms and solutions.