# NVIDIA Holoscan Rover — Edge AI Robot The NVIDIA Holoscan Rover demonstrates the Holoscan SDK — NVIDIA's platform for high-throughput, low-latency AI inference at the edge — applied to a mobile robot platform. The rover uses a Jetson AGX Orin as its compute module and runs a Holoscan pipeline to process sensor data (camera, LIDAR, depth) and drive real-time navigation or manipulation behaviors. Holoscan is NVIDIA's answer to building sensor-processing pipelines that saturate GPU bandwidth: data flows between operators as zero-copy CUDA tensors, enabling sub-millisecond end-to-end latency for safety-critical or real-time robotics applications. ## What Holoscan Adds vs. ROS - Zero-copy CUDA tensor transport between pipeline stages - Deterministic scheduling with nanosecond-resolution timing - Native integration with Jetson NVENC/NVDEC video paths - gRPC-based distributed pipeline across multiple Jetsons ## Hardware - NVIDIA Jetson AGX Orin (64 GB recommended) - Mobile rover chassis with differential drive - OAK-D Pro depth camera or RealSense D435i - NVIDIA A2 or embedded GPU accelerator (optional) ## Source - SDK: https://github.com/nvidia-holoscan/holoscan-sdk - Rover sample: https://github.com/nvidia-holoscan/holohub - License: Apache 2.0 - Docs: https://docs.nvidia.com/holoscan/
Category: Other Robots
| Item | Qty | Notes | |------|-----|-------| | NVIDIA Jetson AGX Orin (64 GB) | 1 | Primary compute for Holoscan | | Mobile rover chassis | 1 | Differential-drive | | OAK-D Pro or RealSense D435i | 1 | Depth + color sensor input | | 4S LiPo battery | 1 | Powers chassis and Jetson via UBEC |