diff --git a/Makefile b/Makefile
index bc429ce..f3caf5f 100644
--- a/Makefile
+++ b/Makefile
@@ -11,12 +11,22 @@ BIN_DIR := $(BUILD_DIR)/bin
 # --- gpr_cpp: external C++/CUDA 3D SAR library (VNA_AlGo). Build it first with
 # `bash $(GPR_CPP_DIR)/build_cmake.sh`; it is NOT built by this Makefile. CUDA
 # headers and libnvrtc come from the pip nvidia-* packages inside the VNA_AlGo
-# venv (same detection as build_cmake.sh); libcuda is the system driver.
-GPR_CPP_DIR ?= /home/awe/Documents/VNA_AlGo/gpr_cpp
+# venv (same detection as build_cmake.sh); when those are absent (e.g. Jetson,
+# where CUDA comes with JetPack) the system toolkit at /usr/local/cuda is used.
+# libcuda is always the system driver.
+GPR_CPP_DIR ?= $(HOME)/Documents/VNA_AlGo/gpr_cpp
 GPR_CPP_LIB ?= $(GPR_CPP_DIR)/build/libgpr_core_cpp.a
 GPR_VENV ?= $(GPR_CPP_DIR)/../.venv
 GPR_CUDA_INC_DIR ?= $(dir $(firstword $(wildcard $(GPR_VENV)/lib/python*/site-packages/nvidia/*/include/nvrtc.h)))
 GPR_NVRTC_LIB ?= $(firstword $(wildcard $(GPR_VENV)/lib/python*/site-packages/nvidia/*/lib/libnvrtc.so*))
+# System-CUDA fallback. An empty GPR_CUDA_INC_DIR would also corrupt the compile
+# command: a bare `-I` swallows the next argument.
+ifeq ($(strip $(GPR_CUDA_INC_DIR)),)
+GPR_CUDA_INC_DIR := /usr/local/cuda/include
+endif
+ifeq ($(strip $(GPR_NVRTC_LIB)),)
+GPR_NVRTC_LIB := $(firstword $(wildcard /usr/local/cuda/lib64/libnvrtc.so*))
+endif
 GPR_CUDA_LIB_DIR := $(dir $(GPR_NVRTC_LIB))
 # libnvrtc from pip ships only as libnvrtc.so.N (no .so symlink), so link the
 # full path; the rpath makes the binary find it without LD_LIBRARY_PATH.
diff --git a/README.md b/README.md
index 595b484..5e3b5cb 100644
--- a/README.md
+++ b/README.md
@@ -77,7 +77,10 @@ controls plus a pixel-identical live plot through an embedded web UI.
 - For the `sar3d` processor (3D SAR): an NVIDIA GPU with the driver installed
   (`libcuda`) and the external `gpr_cpp` library from the VNA_AlGo project. It
   must be built **before** `make` (see below); its location is configurable via
-  the `GPR_CPP_DIR` make variable (default `/home/awe/Documents/VNA_AlGo/gpr_cpp`).
+  the `GPR_CPP_DIR` make variable (default `$HOME/Documents/VNA_AlGo/gpr_cpp`).
+  CUDA headers/libnvrtc are taken from the pip `nvidia-*` packages in the
+  VNA_AlGo venv when present, otherwise from the system toolkit at
+  `/usr/local/cuda` (Jetson/JetPack, apt `cuda-toolkit`).
 
 ## Build and run
 
@@ -87,8 +90,9 @@ From the repository root:
 python3 -m venv .venv
 .venv/bin/python -m pip install --upgrade pip
 .venv/bin/python -m pip install -r requirements.txt
-# 3D SAR library (external, CMake + NVRTC; CUDA headers come from pip nvidia-* packages):
-bash /home/awe/Documents/VNA_AlGo/gpr_cpp/build_cmake.sh
+# 3D SAR library (external, CMake + NVRTC; CUDA headers come from the pip
+# nvidia-* packages in the VNA_AlGo venv, or /usr/local/cuda as a fallback):
+bash ~/Documents/VNA_AlGo/gpr_cpp/build_cmake.sh
 make                     # build the C++ binaries into build/bin/
 ```
 
@@ -114,6 +118,34 @@ Run the desktop GUI:
 `./start.sh --help`. The acquisition device (LibreVNA, Kamil ADC, …) is detected
 automatically from the active config's `radar.model` — there is no per-device flag.
 
+## Jetson (aarch64) bring-up
+
+Ubuntu 22.04 on Jetson has glibc 2.35, and PyQt6 6.8+ ships only
+`manylinux_2_39` aarch64 wheels — `requirements.txt` therefore pins PyQt6 to
+`<6.8` on aarch64 (6.7.1 is the last release with a compatible wheel). With
+that pin the stock `./start.sh` provisioning works; the extra steps are only
+for the CUDA-backed `sar3d` processor:
+
+```bash
+# 1. gpr_cpp against the JetPack system CUDA (no pip nvidia-* packages needed):
+git clone <VNA_AlGo repo> ~/Documents/VNA_AlGo
+GPR_BUILD_PYTHON=OFF bash ~/Documents/VNA_AlGo/gpr_cpp/build_cmake.sh
+# build_cmake.sh falls back to /usr/local/cuda automatically; override with
+# CUDA_INC=... CUDA_LIB=... if the toolkit lives elsewhere.
+
+# 2. copy the replay dataset (for dataset-replay profiles) and point
+#    radar.replay.dataset_dir in the profile at the local copy — the path in
+#    run_config_examples/*.json is machine-specific.
+
+# 3. normal launch; make picks up /usr/local/cuda when the VNA_AlGo venv has
+#    no pip nvidia-* packages:
+./start.sh --profile run_config_examples/run_config_replay_vna.example.json
+```
+
+If linking fails with `cannot find -lcuda`, the driver stub lives at
+`/usr/local/cuda/lib64/stubs/libcuda.so`; on JetPack the real driver is
+`/usr/lib/aarch64-linux-gnu/nvidia/libcuda.so.1` (present after boot).
+
 ## Running without hardware (emulation)
 
 Two ways to run the full pipeline with no radar attached:
diff --git a/requirements.txt b/requirements.txt
index 8630215..e525707 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,7 +3,11 @@ libusb1>=3.1
 pyserial>=3.5,<4
 pyvisa>=1.14
 PyVISA-py>=0.7  # pure-Python VISA backend ("@py"), required by the SN9000 (and any visa_library="@py") path
-PyQt6>=6.6
+# PyQt6 6.8+ ships only manylinux_2_39 aarch64 wheels (glibc >= 2.39); Jetson /
+# Ubuntu 22.04 has glibc 2.35, so pip would fall back to a source build that
+# dies on qmake. 6.7.1 is the last release with a manylinux_2_28 aarch64 wheel.
+PyQt6>=6.6,<6.8; platform_machine == "aarch64"
+PyQt6>=6.6; platform_machine != "aarch64"
 pyqtgraph>=0.13.7
 rpi-hardware-pwm>=0.2.2,<1
 fastapi>=0.110           # embedded web UI (browser view + controls), served in-process
diff --git a/start.sh b/start.sh
index 255b25b..726d617 100755
--- a/start.sh
+++ b/start.sh
@@ -225,14 +225,29 @@ run_privileged() {
 }
 
 ensure_system_dependencies() {
-    if command -v pkg-config >/dev/null 2>&1 && pkg-config --exists libusb-1.0; then
+    # Qt >= 6.5 xcb platform plugin needs libxcb-cursor0; aarch64/Jetson images
+    # (and some minimal desktops) ship without it, and the GUI then dies at
+    # startup with "could not load the Qt platform plugin xcb". apt-only: other
+    # distros/managers name these differently and usually have them preinstalled.
+    local qt_gui_packages_missing=0
+    if command -v apt-get >/dev/null 2>&1 && command -v dpkg-query >/dev/null 2>&1; then
+        local package
+        for package in libxcb-cursor0 libxkbcommon-x11-0; do
+            if ! dpkg-query -W -f='${Status}' "${package}" 2>/dev/null | grep -q "install ok installed"; then
+                qt_gui_packages_missing=1
+            fi
+        done
+    fi
+
+    if command -v pkg-config >/dev/null 2>&1 && pkg-config --exists libusb-1.0 \
+        && ((qt_gui_packages_missing == 0)); then
         return
     fi
 
     echo "[start.sh] Installing system dependency: libusb-1.0 dev headers..."
     if command -v apt-get >/dev/null 2>&1; then
         run_privileged apt-get update
-        run_privileged apt-get install -y pkg-config libusb-1.0-0-dev
+        run_privileged apt-get install -y pkg-config libusb-1.0-0-dev libxcb-cursor0 libxkbcommon-x11-0
     elif command -v dnf >/dev/null 2>&1; then
         run_privileged dnf install -y pkgconf-pkg-config libusb1-devel
     elif command -v yum >/dev/null 2>&1; then