The following general guidelines must be followed throughout all laboratory sessions in ECE-381: Applied Machine Learning. These instructions ensure consistency, reproducibility, and effective utilization of the Jetson Orin Nano platform and associated toolkits.

1. Jetson Orin Nano Setup
   • Ensure your Jetson Orin Nano board is properly connected with keyboard, mouse, and monitor for standard mode, or via USB-C to your laptop for headless mode.
   • For headless access, connect to the board using the command ssh <username>@192.168.55.1  from your terminal or PowerShell.
   • Use the default login credentials provided by the instructor:
     Username: ECE381-<Jetson #>
     Password: machinelearning<Jetson #>
     


2. Docker Environment
   • All labs rely on Docker containers to isolate the machine learning environments. Ensure Docker with NVIDIA runtime is installed and functional.
   • Use provided scripts (e.g., docker_dli_run.sh or container-specific run.sh) to initialize lab environments.
   • For persistent storage of notebooks, data, and outputs, always mount local directories using the -v flag with Docker.


3. JupyterLab and WebUI Access
   • For Labs 1 and 2, access JupyterLab through the browser at 192.168.55.1:8888. The default password is dlinano.
   • For Labs 3 and 4, respective WebUIs will be hosted at ports like 0.0.0.0:7860 (Stable Diffusion) or local URLs printed in terminal. Always open the links in Chrome or Firefox.


4. Camera and Device Usage
   • Ensure the USB webcam is connected before launching Docker containers. If the container cannot detect /dev/video0, restart Docker and reconnect the device.
   • Only one process can access the camera at a time. If errors arise, shut down all active kernels or containers using the camera.


5. SSD and Storage Management
   • For Labs 3b and 4, external SSDs must be mounted at /mnt/nvme/my_storage.
   • Format the SSD using ext4 and configure /etc/fstab to enable auto-mount at boot.
   • Always store large models and generated outputs in the SSD to avoid filling up the Jetson’s internal memory.


6. Model Execution Guidelines
   • Restart Docker containers if you encounter CUDA or memory issues.
   • Always run conversion scripts (e.g., ModelConversion.py) before deploying YOLOv11n models for fast inference.
   • Use GPU-optimized models (TensorRT where applicable) for efficient frame processing and real-time output.


7. Submission and Documentation
   • Each lab must be submitted as a report (one per group unless specified). Include:
     • Task descriptions
     • Screenshots of results
     • Observations and conclusions
     • Challenges and proposed improvements
   • Ensure your group partner’s name is clearly mentioned in the report.
   • Naming Convention: LabX_GroupY_Report.pdf


8. Troubleshooting Tips
   • If your camera freezes in Jupyter or WebUI, shut down the kernel or container, restart it, and rerun all code cells.
   • If Docker throws numpy auto-update errors, rerun the same command as instructed.
   • If you encounter GPU-related failures, check available memory using tegrastats or reboot the Jetson board