2024-03-18 Autonomy Hardware Meeting

History

  • We used to transmit images from the drone to the ground station

    • This never really worked, lots of complications

  • We wanted to by NVIDIA Jetson so autonomy code can run onboard the drone

    • Cost us about $1500, was hard to procure

Issues Experienced With Jetson

  • The cost of the Jetson

    • Because of high cost people are scared to fly the Jetson

    • Getting the Jetson ready for a flight test, we’ve already broke a Raspberry Pi because of too high input voltage, could happen to the Jetson

  • Connectivity

    • On top of buying Jetson, we also have to buy a carrier board ($500)

    • The carrier board has 2 UART ports

      • UART 0: The debugging UART used to help get debug statements when setting up the Jetson (only source of debugging communication).

      • UART 1: UART used for communication purposes. If you need to talk to another device you need UART 1. Unfortunately, UART 1 on our board is busted.

    • UART 1 being broken sucks because we needed UART connection to talk to Pixhawk.

    • Two options from here:

      • We can convert UART 0 from a debugging UART to a general UART. If we do this, then if the Jetson needs to be reimaged (which happens frequently) we have lost all communication to the Jetson and if something goes wrong, we don’t know what happened. We do not want to pursue this option.

      • We can buy a TTL cable to convert a USB connection to a UART connection. We have purchased this cable and confirmed that it works.

    • You need to plug in the USB to the Jetson and the UART connection from the TTL cable to the Pixhawk.

    • In Fall 2023, we requested funding for a new carrier board for our Jetson and received funding.

Looking Forward

  • Daniel (Executive Director) he asked me if there is some middle ground between a Jetson and Raspberry Pi. The answer I gave was the Jetson Nano. These are the 4 reasons why I believe the Jetson Nano is better in the long term for our team.

    • Jetson Nano is around $200

    • Still has CUDA enabled hardware

    • Has I/O already integrated into it (no need of carrier board)

    • We can run LTE on the Jetson Nano

Cost

  • The cost is much more manageable, we can restore it if breaks

  • We can buy more so more members can get hands-on experience

CUDA Hardware

  • It can run the ML model

Input/Output

  • Don’t need to buy extra I/O hardware, already integrated

Running LTE + Airside System

How does LTE work right now.

  • There is a UART connection from the Pixhawk to the Raspberry Pi on the drone.

  • There is a software called MavProxy (runs on the Pi) on the drone that forward telemetry messages through a network connection.

  • MavProxy will read from the UART and route to network socket from which ground station can read.

  • MavProxy can stream telemetry to multiple networks host.

  • Primary reason for running LTE is for better form on ground station to drone communication (better RFD 900’s, XBees).

    • Autonomy decided to capitalize on this and use the network connection.

Why can’t LTE run on Jetson

  • LTE requires UART connection. Uh oh, UART is broken on Jetson.

  • We can combine Jetson and RPi on the drone and consolidate to a single computer that runs both LTE and airside system.

    • Has CUDA enabled hardware to run airside system

    • Also has LTE hat for network communication

  • LTE on Raspberry Pi vs. Jetson - SysInt - WARG (atlassian.net)