Skip to main content

Hardware Option 2: Jetson Orin Nano Edge Kit

Overview​

The Jetson Orin Nano Edge Kit is an NVIDIA-powered embedded system designed for on-device AI inference and robot deployment. This setup prioritizes portability and power efficiency over raw performance, making it ideal for testing real-world robot deployments.

Best For: Students focused on embedded systems, edge deployment, and testing on physical hardware with limited budget.

Total Cost: $500 - $700 USD

What is Edge Computing?​

Edge Computing means running AI models and robot control directly on the robot's onboard computer (the "edge") rather than in a datacenter or cloud. This approach:

  • ✅ Reduces Latency: No network round-trip (critical for real-time control)
  • ✅ Works Offline: No internet dependency
  • ✅ Improves Privacy: Sensor data stays on device
  • ❌ Limited Compute: Smaller models and lower-quality simulations
  • ❌ Power Constrained: Battery life considerations

Hardware Specifications​

Jetson Orin Nano Developer Kit​

ComponentSpecification
GPUNVIDIA Ampere (1024 CUDA cores, 32 Tensor cores)
CPU6-core ARM Cortex-A78AE (2.0 GHz)
RAM8GB 128-bit LPDDR5 (102 GB/s bandwidth)
StorageMicroSD card (64GB+ recommended) + NVMe SSD support
AI Performance40 TOPS (INT8)
Power7W - 15W configurable TDP
Dimensions100mm × 79mm × 31mm
Weight~300g (0.66 lbs)
Price$499 USD (Developer Kit)

Required Accessories​

ItemPurposeCost
Power Supply19V 4.74A (90W) USB-C PDIncluded with Dev Kit
MicroSD Card64GB+ UHS-I (boot drive)$15 - $30
NVMe SSD256GB M.2 2280 (storage expansion)$30 - $50
USB Keyboard/MouseSetup and debugging$20 - $40
HDMI Cable + MonitorInitial configuration$10 - $20 (if not owned)
Ethernet CableNetwork connectivity$5 - $10
USB-C HubAdditional ports (optional)$20 - $40

Total Cost: $599 - $689 USD (with all accessories)

Performance Comparison: Jetson vs Workstation​

TaskJetson Orin NanoRTX 3060 Workstation
Gazebo Simulation10-15 FPS (basic scenes)30-40 FPS
Isaac Sim❌ Not Supported (insufficient VRAM)20-30 FPS
Unity ML-Agents❌ Not Recommended3-4 hours/1M steps
VLA Inference (7B)✅ 0.5-1 token/sec (quantized)2-3 tokens/sec
Object Detection (YOLOv8)✅ 30 FPS @ 640x640120+ FPS
ROS 2 Node Count20-30 nodes max100+ nodes
Power Consumption7-15W400-600W (full system)

Key Takeaway: Jetson is NOT suitable for Isaac Sim due to 8GB RAM and limited GPU memory. Use for deployment testing and inference only.

Setup Instructions​

Step 1: Flash JetPack 6.0 to MicroSD Card​

On Your Host Computer (Windows/Linux/Mac):

  1. Download NVIDIA SDK Manager from developer.nvidia.com/sdk-manager
  2. Install SDK Manager: 
    # On Ubuntu host
    sudo apt install ./sdkmanager_[version].deb
  3. Launch SDK Manager and sign in with NVIDIA Developer account
  4. Select target hardware: Jetson Orin Nano Developer Kit
  5. Select JetPack version: 6.0 (Ubuntu 22.04 base)
  6. Choose installation method: Manual Setup (MicroSD card)
  7. Download components (this may take 30-60 minutes)

Flash MicroSD Card:

# Insert 64GB+ MicroSD card into host computer
# SDK Manager will guide through flashing process

# Alternatively, use Etcher for pre-built image
# 1. Download Jetson Orin Nano image from NVIDIA
# 2. Flash with balenaEtcher to MicroSD card

Step 2: Initial Boot and Configuration​

  1. Insert flashed MicroSD card into Jetson
  2. Connect HDMI monitor, keyboard, mouse, Ethernet
  3. Connect USB-C power supply (19V 4.74A)
  4. Jetson will boot automatically (first boot takes 5-10 minutes)
  5. Complete Ubuntu setup wizard:
    • Username: student (or your preference)
    • Password: Choose strong password
    • Timezone and language settings

Why? MicroSD cards are slow for development. NVMe SSD provides 10-20x faster read/write speeds.

# After first boot, check if NVMe SSD is detected
lsblk | grep nvme

# Expected output if SSD installed:
# nvme0n1     259:0    0   256G  0 disk

# Format and mount NVMe SSD
sudo mkfs.ext4 /dev/nvme0n1
sudo mkdir /mnt/nvme
sudo mount /dev/nvme0n1 /mnt/nvme

# Make mount permanent
echo '/dev/nvme0n1 /mnt/nvme ext4 defaults 0 2' | sudo tee -a /etc/fstab

# Move home directory to NVMe for better performance
sudo rsync -aXS /home/ /mnt/nvme/home/
sudo mv /home /home.bak
sudo ln -s /mnt/nvme/home /home

Step 4: Configure Power Mode​

Jetson supports multiple power modes. For robotics development, use 15W mode for best performance.

# Check current power mode
sudo /usr/sbin/nvpmodel -q

# Set to 15W mode (maximum performance)
sudo /usr/sbin/nvpmodel -m 0

# Verify GPU is active
sudo tegrastats

# Expected output includes:
# GPU@1300MHz

Step 5: Install Development Tools​

# Update system
sudo apt update && sudo apt upgrade -y

# Install essential build tools
sudo apt install -y build-essential cmake git wget curl \
    python3-pip python3-dev python3-venv \
    libopencv-dev libboost-all-dev

# Install ROS 2 dependencies (covered in software-ros2.md)
# Install PyTorch for Jetson
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Use Cases: When to Use Jetson​

✅ Ideal For:​

  1. On-Device Inference

    • Running pre-trained VLA models for robot control
    • Real-time object detection (YOLOv8, DETR)
    • Sensor fusion and localization (EKF, particle filters)

  2. Physical Robot Testing

    • Deploying code to humanoid platforms (Unitree, Boston Dynamics)
    • Testing latency-sensitive control loops
    • Battery-powered autonomous operation

  3. Edge AI Experiments

    • Quantized model deployment (INT8, FP16)
    • Model optimization with TensorRT
    • Power efficiency benchmarking

❌ NOT Suitable For:​

  1. High-Fidelity Simulation

    • Isaac Sim requires 12GB+ VRAM (Jetson has 8GB shared RAM)
    • Gazebo performance limited to basic scenes
    • Unity ML-Agents training too slow

  2. Large Model Training

    • 8GB RAM insufficient for training transformer models
    • Training should happen on workstation or cloud

  3. Multi-Robot Swarms

    • Limited RAM for running 50+ ROS 2 nodes simultaneously

Limitations and Workarounds​

LimitationImpactWorkaround
No Isaac Sim SupportCan't run Module 3 locallyUse cloud instance or workstation for Module 3
8GB Shared RAMLimited concurrent ROS nodesProfile and optimize node memory usage
ARM ArchitectureSome x86 binaries incompatibleUse ARM-compatible containers or compile from source
MicroSD SpeedSlow disk I/OUpgrade to NVMe SSD (highly recommended)
No Display PortHDMI onlyUse HDMI monitor or remote desktop (VNC/NoMachine)

Hybrid Approach: Jetson + Cloud​

Many students use a hybrid setup:

  • Jetson Orin Nano: For Module 1 (ROS 2), Module 4 (VLA inference), and capstone deployment
  • Cloud GPU Instance: For Module 2 (Unity ML-Agents training) and Module 3 (Isaac Sim)
  • Total Cost: $599 (Jetson) + $200-300 (cloud usage) = $799-899 over 13 weeks

This approach provides the best balance of cost, portability, and performance.

Verification Checklist​

After setup, verify your Jetson:

  • JetPack 6.0 installed: cat /etc/nv_tegra_release
  • GPU active: sudo tegrastats shows GPU utilization
  • Power mode set to 15W: sudo nvpmodel -q
  • NVMe SSD mounted (if installed): df -h | grep nvme
  • CUDA available: python3 -c "import torch; print(torch.cuda.is_available())"
  • ROS 2 Humble installed (see software-ros2.md)
  • Network connectivity: ping google.com

Cost Breakdown​

ItemCostNotes
Jetson Orin Nano Dev Kit$499One-time purchase
MicroSD Card 64GB$20Required for boot
NVMe SSD 256GB$40Highly recommended
Accessories (cables, etc.)$40Keyboard, mouse, HDMI
Total$5993-5 year hardware lifespan
Cloud Supplement$200-300For Isaac Sim (Modules 3)
Grand Total$799-899Complete hybrid setup

Next Steps​

  1. Install ROS 2 on Jetson → Software: ROS 2 Setup
  2. Set Up Remote Development → Use VS Code Remote SSH for comfortable coding
  3. Deploy First Robot Node → Test inference latency and performance

Additional Resources​

Jetson Configured? → Continue to ROS 2 Installation