CSE 598 - Action and Perception Laboratory

Author: Justin Lu | lujust@umich.edu

My implementation of the CSE 598 labs that can be run locally on a GPU-equipped computer with CUDA 12+, without having to struggle with Colab and its many, many flaws

Requirements

To run locally, the host computer must have the following installed:

• Docker
• CUDA 12.x
• NVIDIA Container Toolkit

Installation & Usage (Docker)

1. Initialize Git LFS

Robot mesh assets (.obj) are tracked & stored using Git Large File Storage (LFS), so pulling them is required before starting.

# Install Git LFS
sudo apt-get install git-lfs

# Pull assets
git lfs install
git lfs pull

2. Build the Docker image for the lab yo want

Docker service names are defined in the compose .yml files - we'll use lab02 as an example for the remainder of the README.

# First go into the correct directory
cd lab02-playground-intro

#  This will take a while to complete
docker compose -f docker-compose-jupyter.yml build lab02

3. Spin up the Docker Container

# Spin up container
docker compose -f docker-compose-jupyter.yml up lab02

4. Open Jupyter Notebook session in your browser

# You can also paste the URL directly into your browser
start http://localhost:8888

Alternate Usages

Open a shell inside the Container (won't start Jupyter Notebook server)

# Run bash inside container
docker compose -f docker-compose-jupyter.yml run --rm lab02 bash

# (Inside container): now you can manually run main.py
python3 main.py