Duckitown segmentations using Transformers 🤖

This is a repo implementing a visual-servoing agent for lane following and obstable avoidance.

Phase 0: System update

• 💻 Always make sure your Duckietown Shell is updated to the latest version. See installation instructions

• 💻 Update the shell commands: dts update

• 💻 Pull latest containers on your laptop: dts desktop update

• 🚙 Clean and update your Duckiebot: dts duckiebot update ROBOTNAME (where ROBOTNAME is the name of your Duckiebot chosen during the initialization procedure.)

• 🚙 Reboot your Duckiebot.

Phase 1s: Nvidia-Docker setup (optional)

Note: Only do if you have nvidia GPU, otherwise go to the next phase.

• 💻 Set-up nvidia-docker2. See Installation Nvidia-docker2
• 💻 Follow this to change docker runtime, particularly, run the next command:

  sudo tee /etc/docker/daemon.json <<EOF                             
  {  "default-runtime": "nvidia",
      "runtimes": {
          "nvidia": {
              "path": "/usr/bin/nvidia-container-runtime",
              "runtimeArgs": []
          }
      }
  }
  EOF
  sudo pkill -SIGHUP dockerd
  

Phase 2s: Docker setup

• 💻 Download the following docker image using docker pull mikes96/challenge-aido_lf-baseline-duckietown-ml:daffy-amd64

• 💻 Re-tag docker image doing docker tag <IMAGE-TAG> duckietown/challenge-aido_lf-baseline-duckietown-ml:daffy-amd64

• 🤖 Become a transformer!

Phase 3s: Play around

Build your code using:

💻$ `dts exercises build`

You can see the segmentation masks in the simulator by running

💻$ `dts exercises test --sim` 

Sadly, this repo does not run (yet) in the Jetson Nano, nevertheless, you can use your beefy GPU to speed-up inference by running:

💻$ `dts exercises test --duckiebot_name mycroft --local` 

Congratulations!, you are now a 🤖