Courses / theory

Machine Learning Foundations for Robotics

advanced~18 hours15 lessonsEnvironment: pyodide

Prerequisites: python3, maths-for-robotics

Start: The question you're now qualified to ask: learn it, or write it?Practice questions

Modules

Module 1 — Learning, and when not to
  1. 1The question you're now qualified to ask: learn it, or write it?
  2. 2Supervised learning: features, labels, loss
  3. 3Gradient descent from scratch (and why you've met it before)
  4. 4Overfitting, and the robotics-specific version: distribution shift
Module 2 — Neural networks from scratch
  1. 1The neuron, the layer, the forward pass
  2. 2Backpropagation, derived — then implemented in 30 lines
  3. 3The training loop: batches, learning rates, and what divergence looks like
  4. 4Widget: watch a network learn
Module 3 — Learned perception
  1. 1Convolution: a filter you don't have to design
  2. 2Where the model sits in a ROS 2 stack (and the latency budget it must respect)
  3. 3Project: a tiny image classifier, in NumPy, in your browser
Module 4 — Learning to act
  1. 1Learned dynamics: when the physics is too ugly to write
  2. 2Imitation learning: behavior cloning, and its famous flaw
  3. 3Project: behavior-clone a wall-follower — then watch it fail
  4. 4Sim-to-real: the reality gap, and the sentence to memorize

Textbooks

  • Pattern Recognition and Machine Learning — Christopher M. Bishopbook
  • Deep Learning — Goodfellow, Bengio & Courvillefree-book
    Free online from the authors