OpenVINO#
OpenVINO is the primary toolkit for optimizing and deploying deep-learning inference in Robotics AI Suite applications. It supports models from common frameworks and can target available Intel compute devices.
Use the current OpenVINO installation documentation for the selected environment.
Reference Applications#
OpenVINO reference applications cover object detection, segmentation, and RealSense camera workflows.
Run semantic segmentation on RealSense image data using OpenVINO inference.
Deploy object-detection workloads with ROS 2 camera inputs and OpenVINO acceleration.
Process multiple camera streams in a single OpenVINO-powered demo pipeline.
Use a YOLOv8 model with OpenVINO for accelerated object detection on robotics systems.
Optimize and deploy perception, manipulation, and vision-language-action models with OpenVINO.
Convert, compress, benchmark, and validate the Pi0.5 vision-language-action model.
Additional Guidance#
OpenVINO model guidance includes reusable perception, manipulation, and foundation-model guidance. The workflows require the platform getting-started guide when used with the Humanoid Toolkit.
Benchmarking#
Use the upstream OpenVINO Benchmark Tool to estimate deep-learning inference throughput and latency on supported Intel® devices. Install OpenVINO and its samples with the OpenVINO sample guidance before benchmarking.
Use the same OpenVINO version to convert a model and to run inference unless the model’s documentation explicitly supports a different compatibility path.