Isaac ROS FoundationStereo
Type: Model / ROS Package Tags: NVIDIA, Isaac ROS, FoundationStereo, stereo depth, disparity, robotics, perception, CUDA, TensorRT, Jetson Related: NVIDIA-Isaac-ROS, NVIDIA-Isaac-for-Mobility, NVIDIA-Isaac-for-Manipulation, Isaac-ROS-DNN-Stereo-Depth, Isaac-ROS-DNN-Inference, Isaac-ROS-NITROS, Isaac-ROS-Visual-SLAM, Isaac-ROS-nvblox, NVIDIA-Jetson-Platform, NVIDIA-Jetson-Thor, TensorRT, NVIDIA-VPI Sources: https://nvidia-isaac-ros.github.io/concepts/stereo_depth/foundationstereo/index.html, https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_dnn_stereo_depth/isaac_ros_foundationstereo/index.html, https://nvidia-isaac-ros.github.io/releases/index.html Last Updated: 2026-04-29
Summary
Isaac ROS FoundationStereo is NVIDIA’s Isaac ROS support for FoundationStereo, a foundation model for stereo depth estimation. It predicts disparity from stereo RGB image pairs and is useful when a robot needs depth from stereo cameras for mapping, reconstruction, perception, or manipulation.
Detail
Purpose
Stereo depth is a core input for mobile robots and manipulation systems. FoundationStereo provides a deep learning approach to disparity estimation that can generalize across scenes, camera types, lighting variation, occlusions, and non-standard camera parameters better than many classical stereo methods.
Key capabilities
- Stereo disparity prediction from RGB stereo image pairs.
- Transformer-based deep stereo approach trained for broad generalization.
- FoundationStereo v2 support surfaced in current Isaac ROS release notes.
- Isaac ROS package and tutorials for running the model in ROS 2 pipelines.
- Best suited to applications where high-quality depth matters more than strict real-time performance.
- Depth output can feed Isaac-ROS-nvblox reconstruction, NVIDIA-Isaac-for-Mobility navigation, and NVIDIA-Isaac-for-Manipulation obstacle perception.
NVIDIA context
FoundationStereo sits at the camera-perception edge of Isaac ROS. Its stereo-derived depth can become robot-world geometry through Isaac-ROS-nvblox, which then supports mobile navigation or manipulation planning on NVIDIA-Jetson-Platform and other NVIDIA GPU systems.
Connections
- NVIDIA-Isaac-ROS - parent ROS 2 package ecosystem for FoundationStereo.
- NVIDIA-Isaac-for-Mobility - AMR workflow where stereo depth can support mapping and navigation.
- NVIDIA-Isaac-for-Manipulation - manipulation workflow where stereo depth can feed obstacle perception.
- Isaac-ROS-DNN-Stereo-Depth - parent Isaac ROS stereo-depth package family that includes FoundationStereo.
- Isaac-ROS-DNN-Inference - DNN infrastructure for deep stereo disparity model execution.
- Isaac-ROS-NITROS - accelerated transport layer for disparity image and tensor data.
- Isaac-ROS-Visual-SLAM - adjacent stereo-camera mobility component.
- Isaac-ROS-nvblox - reconstruction layer that can consume depth generated from stereo cameras.
- NVIDIA-Jetson-Platform - robot-edge deployment family for Isaac ROS perception.
- NVIDIA-Jetson-Thor - current-generation Jetson platform surfaced in recent Isaac ROS release notes.
- TensorRT - inference optimization path for deep learning models in Isaac ROS.
- NVIDIA-VPI - adjacent accelerated vision library for stereo and image processing on Jetson.
Source Excerpts
- NVIDIA docs describe FoundationStereo as a foundation model for stereo depth estimation.
- The model predicts per-pixel disparity from stereo camera image pairs.
- Isaac ROS release notes include FoundationStereo and FoundationStereo v2 support in current 4.x releases.