seb/saltylab-firmware
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Page: Compute Platform
Pages
Compute Platform FSESC 4.20 Plus Hub Motor Measurements Project Status Roadmap SIM7600X 4G Cellular UWB Follow-Me System Vision System
Clone
1
Compute Platform
seb edited this page 2026-02-28 20:07:27 -05:00
Unescape Escape
Table of Contents
This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

Compute Platform

Current: Jetson Nano (being replaced)

Spec Value
GPU 128-core Maxwell
AI Performance 0.47 TOPS
CPU 4× Cortex-A57 @ 1.43GHz
RAM 4GB LPDDR4
Storage microSD
Power 5-10W
JetPack 4.6 (Ubuntu 18.04)
Status Retiring — replaced by Orin Nano

Incoming: Jetson Orin Nano Super Developer Kit (arriving March 1)

Spec Value
Model NVIDIA Jetson Orin Nano Super Developer Kit
GPU 1024-core Ampere
AI Performance 67 TOPS (140× improvement over Jetson Nano)
CPU 6× Cortex-A78AE @ 1.5GHz
RAM 8GB LPDDR5
Storage 1TB NVMe M.2 2242 (Timetec PCIe Gen3×4, 2100/1600 MB/s)
Power 7-25W (configurable)
USB 3× USB3
CSI 4× MIPI CSI-2 lanes
WiFi Built-in (on carrier board)
JetPack 6.x (Ubuntu 22.04)
Form Factor 70×45mm module (same as Jetson Nano)
Price $371.41 CAD (Amazon)

What This Means for the Team

Performance Gains

  • Person detection/tracking: 20 FPS → real-time multi-stream
  • SLAM: Can run at higher resolution and rate
  • On-board LLM: Small models (Llama 3B) now feasible
  • Voice: Local Whisper transcription possible
  • Multi-camera: Enough USB bandwidth + compute for RealSense + additional cameras

Software Changes Required

  • JetPack 6 (Ubuntu 22.04) — ROS2 Humble native support (no more building from source)
  • Docker base images need updating from l4t-base:r32.x → l4t-base:r36.x
  • TensorRT models need re-export for Ampere architecture
  • Boot: NVMe (not microSD) — flash via NVIDIA SDK Manager
  • CSI cameras: 4 lanes available if we add onboard cameras later

Migration Tasks

  1. Flash JetPack 6 onto NVMe SSD via SDK Manager (from Ubuntu host or VM)
  2. Install ROS2 Humble (apt, not source build)
  3. Re-test RealSense + RPLIDAR drivers (should work, may need updated packages)
  4. Re-export TensorRT models (SSD-MobileNet-v2 for person tracking)
  5. Update docker-compose.yml base images
  6. Update saltybot_bridge serial port path if different (/dev/ttyACM0)
  7. Benchmark inference FPS and update SLAM parameters (can likely increase from 2Hz)

Pin Compatibility

The Orin Nano Dev Kit carrier board has different I/O than the Jetson Nano Dev Kit:

  • USB: USB-C (power + data) + USB3 Type-A ports
  • HDMI: Yes (for debug)
  • GPIO: 40-pin header (compatible)
  • UART: Available on GPIO header for FC communication
  • No barrel jack — USB-C power only (5V/3A minimum)

FC ↔ Jetson Wiring (unchanged)

Jetson UART        Drone FC (UART1)
──────────         ────────────────
TX (Pin 8)   ──→  RX (PA10)
RX (Pin 10)  ──→  TX (PA9)
GND          ──→  GND

Battery / Power Notes

  • Orin Nano needs 5V via USB-C — use a 36V→5V DC-DC converter from main battery
  • Max draw ~25W at full AI load — size the DC-DC accordingly (5V/5A minimum)
  • The old Jetson Nano's barrel jack 5V/4A supply won't work — must be USB-C