RobotGuide

Foundations · math

Build the math tools
one by one.

Before touching any algorithm, build the math tools robotics uses, one by one: a group 0 ramp for readers starting from high school (six bridges out of trig identities, derivatives, the dot product, the normal distribution, circular motion, and complex multiplication), plus the probability–optimization axis behind reinforcement learning (probability axioms, conditional expectation, concentration inequalities, fixed points, Markov chains, Bellman, duality, Robbins–Monro) and the contact–convex-geometry axis behind manipulation and tactile sensing (Hertzian contact, the friction cone, force closure, rolling kinematics, slip detection). Every article leads with motivation, proves everything, and runs the numbers.

The bar: after reading every proof on this level, no step inside the lesson pages (the /k/ articles) is left dangling with «why is this allowed». Group 0 is the ramp for high-school starts; the other groups do not depend on it — read both orderings and pick yours.

Every article has the same shape

Motivation first (where you get stuck without it) → two-layer definitions (intuition + rigor, and what the rigorous layer guards against) → complete proofs («easy to see» is banned) → an example you can compute by hand → numbers you can run → where it shows up in robotics → self-test questions.

How this relates to the lesson pages

The /k/ lesson pages tell you which math tool each algorithm leans on; the proofs of those tools live here. When a lesson step feels unjustified, the justification is an article on this level.

Group 0 · starting from high school

From angle-sum identities, derivatives, the vector dot product, the normal distribution, circular motion, and complex multiplication — six things you already know from high school, six bridges into deep water, with notation kept to a minimum

T1 The Matrix Is a Ledger for RotationYou already know that the derivative of f(x) = x² is 2x — and that sin(a+b) and cos(a+b) exist. Arrange them into a 2×2 table and you get the thing that appears ten thousand times in robotics: the rotation matrix. ≈ 25 min2 theorems3 self-test T2 From Slope to Gradient: The First Line of Every Robot Learning AlgorithmYou already know that the derivative of f(x) = x² is 2x. Generalize the «one-dimensional slope» to many dimensions and you get the most frequent symbol in robot learning algorithms: ∇. ≈ 25 min1 theorems3 self-test T3 Vectors and forces: the physical reading of the dot productThe planar dot product a·b = |a||b|cosθ is high school material. In robotics, every «force and motion» problem starts with this one line. About 20 min2 theorems3 self-test T4 Two Rulers and a Better RulerMean, variance, and the normal distribution are high-school statistics. Using only these, this article derives the core of the Kalman filter: two imprecise measurements, combined, can be more accurate than either alone. about 25 min1 theorems3 self-test T5 The Ledger of One Motor Revolution: Radians, Speed, and PowerUniform circular motion, angular velocity, and radians are high-school physics. Every motor nameplate in a robot is a word problem applying these few formulas. about 20 min2 theorems3 self-test T6 Complex multiplication is rotation: the one-dimensional version of quaternionsYou learned complex multiplication in high school: (1+i)(1−i) = 2. It hides one of the most valuable mathematical facts in the robot industry: multiplying by a complex number = rotating through an angle. About 20 min2 theorems3 self-test

Linear algebra & geometry

Basis and coordinates, matrix multiplication as transform chaining, Jacobians, eigenvectors, SVD and the pseudoinverse, Taylor expansion — the second foundation of robotics: how coordinates are kept, how transforms chain, how approximations get an error receipt

T7 Bases and coordinates: why robots put everything in matricesThe same point reads differently under a different basis — the vector is objective; coordinates are bookkeeping. 45 min3 theorems4 self-test T8 Matrix Multiplication: Chaining Transformations, Not a Table OperationThe j-th column of AB = A applied to b_j — matrix multiplication is the pipeline «first B, then A». The order is a physical fact: rotate-then-stretch ≠ stretch-then-rotate. About 40 min3 theorems4 self-test T9 The Jacobian: Rotate a Joint One Degree — How Far Does the End Effector Move?The Jacobian is the conversion table for «each joint moves a little — how much does the end effector follow» — the multivariable chain rule arranged as a matrix. Velocity control, force control, singularity analysis: all of it is read off this one table. About 45 min3 theorems4 self-test T10 Eigenvectors: The Directions a Matrix Dare Not TouchA matrix skews most directions, but a few directions only get stretched or shrunk — those are the eigen-directions, and the scaling factors are the eigenvalues. The principal axes of a point cloud, the principal modes of a vibration, the dominant timescales of a system — all of them. about 45 min3 theorems4 self-test T11 SVD and the Pseudoinverse: When There Is No Exact Solution, Find the Nearest OneAny matrix = rotate → stretch along orthogonal axes → rotate again. When the equations have no solution (overdetermined), least squares gives the nearest solution; when a matrix is singular (ill-conditioned), the pseudoinverse gives the most robust one. Robot IK and calibration do these two things every day. about 50 min3 theorems4 self-test T12 Taylor expansion: putting a straight line to work in place of a curveNear a point, a smooth curve = a straight line (first order) + a curvature correction (second order) + even smaller corrections... The Taylor expansion attaches an error receipt to every «approximation». Robot linearization, Newton's method, error analysis — all of it is this. About 40 min3 theorems4 self-test

Probability & expectation

Axioms, construction, conditional expectation, concentration inequalities — the full groundwork under the word «expectation» in RL

T13 Probability spaces: why «random» needs axiomsThree answers to the same problem means not that math failed, but that «random» was never defined. Starting from Bertrand's paradox, this piece builds the sample space, the event collection, and the probability measure, one by one. 50 min3 theorems4 self-test T14 Expectation: A Rigorous Construction of «Average»«Average reward» is trivial to compute in the discrete finite case, but continuous or infinite cases demand a construction. Lebesgue's three-step construction gives expectation a rigorous definition — and along the way explains why linearity needs no independence. About 55 min3 theorems4 self-test T15 Conditional Expectation: What the Vertical Bar Really MeansE[Y|X] is not a number; it is a random variable — «the part of Y that X can explain». The tower property and the projection property are where the legitimacy of TD, GAE, and least-squares value-function approximation all comes from. ≈ 60 min2 theorems4 self-test T16 Four inequalities: the quantitative answer to «how many samples are enough»Markov uses only non-negativity, Chebyshev adds variance, Hoeffding adds independence — tighter layer by layer, with more premises layer by layer. How many episodes policy evaluation needs has a formula. 45 min4 theorems4 self-test

Convergence & fixed points

Where «it certainly converges» and «time average equals ensemble average» actually come from

T17 Norms and fixed points: where «it always converges» comes from«Value iteration always converges» is not an empirical fact; it is a corollary of Banach's fixed point theorem. The Bellman optimality operator is a γ-contraction in the sup norm — γ<1 is the entire premise, and at γ=1 every guarantee vanishes. 55 min2 theorems4 self-test T18 Markov Chains: When the Time Average Equals the Ensemble AverageWhy is it legitimate for the replay buffer to destroy temporal correlations, and what exactly is the on-policy distribution? The answers live in the stationary distribution and the ergodic theorem — provided the chain is irreducible and aperiodic. You need both; neither is optional. about 55 min2 theorems4 self-test

Decision & optimization

Markov axioms, dynamic programming, constrained optimization

T19 The Mathematical Construction of the MDP: What One Axiom Is WorthAll of reinforcement learning rests on a single axiom: the future depends only on the current state. Bellman's principle of optimality can be proven rigorously by contradiction — provided the «state» really does contain all the relevant information. about 50 min2 theorems4 self-test T20 Dynamic Programming: The Accounting for Two IterationsValue iteration is cheap per step but needs many steps; policy iteration is expensive per round but needs few rounds. The policy improvement theorem guarantees strict non-decrease each round — which one is faster is a computable question on any concrete problem. ≈ 50 min2 theorems4 self-test T21 Optimization: why constraints cannot rely on penalties aloneOn a convex function the local optimum is the global optimum — that is not a comfort, it is a theorem. TRPO writes a KL constraint instead of a large penalty; the reason lives in duality theory: a penalty coefficient must be hand-picked, a constraint does not. About 55 min3 theorems4 self-test

Stochastic approximation

Iterating with noise — the last foundation block under TD and Q-learning

T22 Stochastic Approximation: Every Term in the Learning-Rate FormulaEvery condition for x_{n+1} = x_n + a_n(observation − x_n) to converge corresponds to a class of blow-up: a step size that never shrinks jitters forever; one that shrinks too slowly drags a noise tail. The two Robbins-Monro conditions are the final word on the learning rates of TD and Q-learning. about 55 min2 theorems4 self-test

Contact & convex geometry

Hertzian contact, the friction cone, force closure, rolling kinematics, slip detection — the five math foundations of manipulation and tactile sensing

T23 The Mathematics of Contact: Hertz TheoryPress a finger into a desk and the force is not a linear function of indentation — it goes as the 3/2 power of indentation. Starting from the part elasticity can fully prove, this article pushes the power law out inch by inch. about 45 min4 theorems4 self-test T24 Friction and the Friction Cone: The Geometry of ConstraintsWhether something slips does not depend on how big the force is — it depends on the direction. Coulomb friction draws a cone in force space, and stuffing that cone into an optimizer means first slicing it into a polytope — how large is the error and how does it converge? This page gives the full answer. about 40 min3 theorems4 self-test T25 Force Closure and Grasp Metrics: ε Is a Linear Program«Grips firmly» can be defined rigorously, and it can be computed into a number: the radius ε of the ball inscribed in the wrench set. How much a two-finger pinch falls short of a three-finger 120° grasp, how much ε drops when you move a finger by 40°, what a friction coefficient is worth in money — all of it is the output of this one LP. About 50 min3 theorems4 self-test T26 Contact Kinematics: Why Rolling Is Not «Sliding Along»A finger rolls 30° over an object's surface — how far does the contact point travel on each side? A ball rolls around a square back to its start — how much has its orientation drifted? Rolling is governed by a set of first-order equations, and the circular path even has a closed-form solution. ≈ 45 min3 theorems4 self-test T27 Slip detection: from «I felt it» to «I computed it»Slip is not an instantaneous event: the tangential force «pre-slips» at the edge of the contact first, and the approach of the ratio |f_t|/f_n toward μ is a measurable precursor. Turning the precursor into a detector is a hypothesis-testing problem — and the optimal detector is given by Neyman-Pearson. About 50 min3 theorems4 self-test

Order and boundaries

Group 0 is the ramp for readers starting from high school — only T1–T6. The other groups assume calculus and linear algebra are in place; the numbering is the dependency order, and skipping ahead is at your own cost. Every article ends with a «boundaries» section that states honestly where it stops and where to go for the rest. Want these tools running end to end on a real task? Head to the roadmap or the labs.