NVIDIA Isaac for Manipulation

Type: Reference Architecture Tags: NVIDIA, Isaac, Isaac ROS, robotics, manipulation, robot arms, motion planning, CUDA, cuMotion, Jetson Related: NVIDIA-Isaac, NVIDIA-Isaac-ROS, Isaac-ROS-cuMotion, Isaac-ROS-nvblox, Isaac-ROS-FoundationPose, Isaac-ROS-FoundationStereo, Isaac-ROS-DNN-Stereo-Depth, Isaac-ROS-DNN-Inference, Isaac-ROS-Object-Detection, Isaac-ROS-Image-Segmentation, Isaac-ROS-Image-Pipeline, Isaac-ROS-AprilTag, Isaac-ROS-NITROS, NVIDIA-Isaac-Sim, NVIDIA-Isaac-Lab, NVIDIA-Jetson-Platform, TensorRT Sources: https://nvidia-isaac-ros.github.io/reference_workflows/isaac_for_manipulation/reference_architecture.html, https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_cumotion/index.html, https://nvidia-isaac-ros.github.io/releases/index.html Last Updated: 2026-04-29

Summary

NVIDIA Isaac for Manipulation is the current Isaac ROS reference architecture for perception-driven robot-arm manipulation. It brings together GPU-accelerated perception, depth processing, 3D reconstruction, object pose estimation, robot configuration, and Isaac-ROS-cuMotion planning so a robotic arm can perform collision-aware manipulation tasks. Current Isaac ROS release notes identify this workflow family as the current name for the workflows formerly known as Isaac Manipulator.

Detail

Purpose

Isaac for Manipulation gives robotics teams a validated starting architecture for building accelerated pick-and-place, inspection, object-following, and contact-rich manipulation systems. The workflow is designed to run with real cameras and robots, with simulated robots in NVIDIA-Isaac-Sim, and with deployment targets such as Jetson Orin or Ampere-and-newer NVIDIA GPU systems.

Key capabilities

Current naming

  • Isaac ROS 4.x documentation uses Isaac for Manipulation for this current reference-workflow area.
  • Earlier public material may use Isaac Manipulator; keep that as an alias in search context, but keep this page as the canonical wiki page.
  • Individual tutorials, bringup packages, orchestration packages, and robot-specific examples should be folded into this page or into the relevant durable component page instead of becoming separate wiki pages.

NVIDIA context

Isaac for Manipulation is where NVIDIA’s robotics deployment stack becomes an integrated manipulation workflow. It combines NVIDIA-Isaac-ROS packages, TensorRT-optimized perception models, NVIDIA-Isaac-Sim simulation, NVIDIA-Isaac-Lab policy workflows, and NVIDIA-Jetson-Platform deployment hardware into one robot-arm architecture.

Connections

Source Excerpts

  • Current Isaac ROS docs describe Isaac for Manipulation as GPU-accelerated libraries and packages for perception-driven manipulation with robotic arms.
  • The reference architecture lists visual input, environment perception, robot configuration, goal-state estimation, motion planning, and hardware platform blocks.
  • Isaac ROS release notes identify Isaac for Manipulation as the current name for workflows formerly known as Isaac Manipulator.

Resources