AUP Teaching Labs

To remove the barriers to teaching.

The AMD University Program (AUP) Teaching Solutions initiative empowers educators with high-quality, hands-on curriculum resources for modern AI and robotics, designed for AMD GPU acceleration. These labs run on AUP Learning Cloud — a JupyterHub platform delivering ROCm-accelerated course environments.

Branches

• main — reimplementation-ready course code: Jupyter notebooks, assets, and the matching Dockerfile for each lab.
• doc — the GitHub Pages web portal (index.html, assets/, per-lab pages). Browse it at https://amdresearch.github.io/aup-teaching-labs/.

Repository Structure

aup-teaching-labs/
└── projects/
    ├── CV/                                  # Computer Vision notebooks
    ├── DL/                                  # Deep Learning notebooks
    ├── LLM/                                 # LLM from Scratch notebooks
    └── Physical-AI/
        ├── Physical-Simulation/
        │   ├── Genesis-Simulation/          # Genesis (formerly PhySim) + Dockerfile
        │   └── Mujoco-Simulation/
        │       ├── mujoco-torch/            # MuJoCo + PyTorch labs + Dockerfile
        │       └── mujoco-MJX/              # MuJoCo MJX labs + Dockerfile
        └── Real-Deployment/
            ├── Robot-Policy-Deployment/     # (planned)
            └── ROS2-Deployment/             # (planned)

Each lab folder contains its Jupyter notebooks, any required assets, and a Dockerfile describing the environment needed to run it.

Course Taxonomy

1. Physical AI

1-1 Physical Simulation

• 1-1-1 Genesis Simulation (projects/Physical-AI/Physical-Simulation/Genesis-Simulation/) Robotics and physics simulation with Genesis on AMD GPUs — load robots into scenes, apply PD controllers, pick-and-place with Inverse Kinematics, and scale to parallel environments.

• 1-1-2 Mujoco Simulation

  • mujoco-torch (.../Mujoco-Simulation/mujoco-torch/) — Robosuite/MuJoCo with PyTorch: controllers and cameras, Gymnasium and rewards, behavior cloning, SmolVLA fine-tuning, PPO, and cross-domain RL.
  • mujoco-MJX (.../Mujoco-Simulation/mujoco-MJX/) — MuJoCo MJX: MJCF concepts, rendering and contacts, control and IK, MuJoCo→MJX, parallel rollouts and domain randomization, and MuJoCo Playground PPO.

1-2 Real Deployment

• 1-2-1 Robot Policy Deployment — placeholder for upcoming content.
• 1-2-2 ROS2 Deployment — placeholder for upcoming content.

Other Labs

• Computer Vision (projects/CV/) — classification, detection, segmentation, tracking, and generative vision models in PyTorch.
• Deep Learning (projects/DL/) — classical ML through neural networks, CNNs, GANs, and Transformers from first principles.
• Large Language Model from Scratch (projects/LLM/) — PyTorch fundamentals to a working LLaMA-style decoder.

Running on AUP Learning Cloud

These notebooks are designed to run on AUP Learning Cloud, AMD's JupyterHub platform for hands-on AI education, which provides pre-built course images with AMD GPU acceleration via ROCm. Each lab's Dockerfile mirrors the image used to run that lab.

Full platform documentation: https://amdresearch.github.io/aup-learning-cloud/

Acknowledgments

Lab content is developed and maintained as part of the AMD University Program, in collaboration with university partners including National Taiwan University (CV, DL) and Nanjing University (LLM).

License

Lab notebooks retain the copyright and license terms from the original AUP Learning Cloud project. See individual notebook headers for details.