Holoscan Platform

A practical guide to the Holoscan Platform, Holoscan Sensor Bridge, and Holoscan SDK.

NVIDIA Holoscan is a comprehensive platform for accelerating AI and high-performance computing applications in healthcare, automotive, and robotics. Use this guide to connect sensors, bring up the platform, run demos, measure performance, and extend Holoscan applications.



Start with the hardware connection guide, then follow the bring-up and demo pages.

Explore the guide

Start here
Hardware Connection
Connect the Jetson, Sensor Bridge, cameras, and network equipment.
Begin here if the hardware is not configured yet.
Setup
Bring Up Process
Prepare JetPack, Docker, the NVIDIA runtime, and the sensor bridge source.
Follow this after the physical connections are complete.
Task
Running Demo
Run the camera capture and display example on the target platform.
Use this for the first end-to-end video test.
SDK
Holoscan SDK Introduction
Learn how Holoscan applications use graphs, operators, and accelerated libraries.
Read this path when developing a custom Holoscan application.

Key Components of Holoscan

  • Hardware: GPUs, FPGAs, Jetson platforms, and sensor interfaces handle intensive real-time workloads. Holoscan Sensor Bridge boards are provided by Lattice Semiconductor, Microchip, and NVIDIA, with the goal of delivering sensor data directly into host-device memory.
  • Software: The Holoscan SDK provides data ingestion, preprocessing, model inference, visualization, Python, C++, and Graph Composer support.

Holoscan Sensor Bridge

The Sensor Bridge links high-bandwidth sensors such as cameras, LiDAR, and medical imaging devices to AI pipelines with low-latency ingestion, synchronized processing, and edge scalability. The current project focus is experimental camera support.

  • Sensor integration: Supports camera workflows and protocols such as GMSL2, GMSL3, and Ethernet.
  • Low-latency ingestion: Uses DPDK-compatible Ethernet where available and preserves temporal accuracy for applications such as surgical guidance and AR overlays.
  • Processing: Synchronizes multi-sensor inputs and supports filtering, normalization, and compression before inference.
  • Edge scalability: Runs on NVIDIA Jetson Orin and IGX Orin platforms, from single-sensor systems to distributed multi-input setups.

Contents Pin