Edge Deployment Skill

Overview

Expert skill for optimizing and deploying machine learning models on robot edge devices including NVIDIA Jetson and embedded systems.

Capabilities

• Configure TensorRT optimization for NVIDIA Jetson
• Set up ONNX model conversion and optimization
• Implement INT8 and FP16 quantization
• Configure DeepStream for video analytics
• Set up CUDA graph optimization
• Implement model pruning and distillation
• Configure DLA (Deep Learning Accelerator) deployment
• Set up multi-stream inference
• Implement ROS2 inference nodes
• Profile and benchmark on target hardware

Target Processes

• nn-model-optimization.js
• object-detection-pipeline.js
• rl-robot-control.js
• field-testing-validation.js

Dependencies

• TensorRT
• ONNX Runtime
• NVIDIA Jetson SDK
• DeepStream

Usage Context

This skill is invoked when processes require deploying ML models on edge devices with optimized inference performance.

Output Artifacts

• TensorRT engine files
• ONNX optimized models
• Quantization configurations
• DeepStream pipeline configs
• Inference benchmark reports
• ROS2 inference node implementations