Field notes

Sensor fusion, perception, path planning, ISO 26262 functional safety, redundancy, V2X and fleet management on NVIDIA Jetson hardware.

Isaac ROS is NVIDIA's toolkit for building robots: pre-built, GPU-accelerated packages for visual SLAM, spatial mapping, object detection and motion planning.

Stereo depth, LiDAR processing, visual SLAM, Gaussian Splatting and Isaac ROS on NVIDIA Jetson, with measured frame rates for Orin and AGX Thor.

Secure boot and fuse programming, OP-TEE and fTPM, container hardening, LUKS encryption, signed OTA updates and zero-trust access on NVIDIA Jetson.

Detection, segmentation, pose and multi-camera analytics on Jetson Orin, with benchmarks and the TensorRT, DeepStream and VPI settings behind them.

Triton on Jetson, model ensembles, A/B testing, hot-swapping, federated learning and edge-cloud routing: production patterns for edge AI inference.

Dynamic shapes, IPluginV3 plugins, INT8 calibration, 2:4 sparsity and Nsight profiling for TensorRT inference on NVIDIA Jetson AGX Orin.

A BiLSTM reads a sequence forward and backward. How to build one in PyTorch or TensorFlow and run it on NVIDIA Jetson with TensorRT.

From PyTorch model to TensorRT engine on Jetson: quantization, pruning, distillation, memory, batch sizing and DLA offload, with benchmarks.

Multi-stream pipelines, custom GStreamer plugins, NVDEC and NVENC, Kafka and MQTT output, Triton and Kubernetes deployment for NVIDIA DeepStream.