Just enough math

Robotics is linear algebra wearing a hard hat. Rotations are matrices, positions are vectors, and a policy is a function you push gradients through. You need to be comfortable, not expert.

Do this

• Essence of Linear Algebra (3Blue1Brown). Watch all of it first. It's the fastest route to intuition and it's free.
• 18.06 Linear Algebra (MIT OpenCourseWare, Gilbert Strang). Lectures 1 to 10 and 21 to 22, with the problem sets. Strang teaches it the way engineers use it.
• Mathematics for Machine Learning (Deisenroth, Faisal, Ong; free PDF). Chapters 2 to 5 for the algebra and calculus you'll meet in training, chapter 6 for probability. Read with a notebook open.

Build

In NumPy, write functions that build a 3D rotation matrix from an axis and an angle, compose two rotations, and apply them to a point. Check them against a library. Then rotate a cube in matplotlib. You've just written the heart of a forward-kinematics chain.

Skip

• Proofs, unless you enjoy them.
• Anything on eigenvalues beyond the idea, until module 06 needs it.

Done when: you can explain what a rotation matrix does, multiply matrices by hand for small cases, and take a gradient of a simple function without panic.