This half-day tutorial addresses the emergence of machine learning techniques, specifically reinforcement learning, within the classic control domain for image-based and non-image based applications. Segway-like platforms and autonomous driving scenarios serve as illustrative models, providing additional insight of learning-based control approaches.

The hands-on component starts with a code walk-through, to modify and experiment with simulator-based learning algorithms. Subsequently, the simulator-trained models are then deployed on 1:10 scale rovers for testing and evaluation. Furthermore, selected topics such as performance metrics, generalization, and architectural design / debugging concepts will be covered as time allows.

Participants are encouraged to form 2-4 person groups for an end of session competition.

• Fundamentals of PyTorch, Python and Numpy
• Fundamentals of Machine Learning / Control Theory / Image Processing
• An i5+ class notebook with:
  • Anaconda / Python 3.5+ and Pandas
  • PyTorch and Jupyter
  • OpenAI Gym including both its run-time and its source code
• Frequently asked Q/A could be found on this page