Mathematical Foundations for Robotics

The study of modern robotics demands a rigorous mathematical framework that enables precise modeling, estimation, and control of mechanical systems operating in uncertain environments. This chapter establishes the essential mathematical foundations that permeate every aspect of contemporary robotics, from the fundamental kinematics that describe robot motion to the sophisticated algorithms that enable autonomous navigation and manipulation. The four mathematical domains presented here—linear algebra, probability theory, optimization theory, and information theory—constitute the analytical toolkit that every robotics practitioner must command. These disciplines are not merely abstract mathematical constructs; they are practical instruments that solve real problems in robot perception, planning, and control.

The interconnection among these mathematical domains becomes evident when we consider the full pipeline of a robotic system. A robot must perceive its environment through noisy sensors, build and maintain a representation of the world, plan trajectories that avoid obstacles while optimizing some objective, and execute control actions that track desired motions despite model uncertainties. Each stage of this pipeline draws upon multiple mathematical disciplines simultaneously. State estimation combines linear algebra for system modeling with probability theory for handling uncertainty. Motion planning relies on optimization theory to find feasible trajectories and linear algebra to represent geometric transformations. Simultaneous Localization and Mapping (SLAM) synthesizes all four domains into a coherent framework for robot navigation in unknown environments.

Linear Algebra

Linear algebra provides the language and computational framework for robotics. The positions, orientations, and velocities of robot components are represented as vectors, while the transformations relating different coordinate frames and the dynamics governing robot motion are expressed as matrices. A thorough understanding of linear algebra—encompassing vector spaces, matrix operations, decompositions, and eigenvalue analysis—is prerequisite to all subsequent topics in this book.

Vector Spaces and Linear Transformations

A vector space (also called a linear space) over a field $\mathbb{F}$ is a set $V$ equipped with two operations: vector addition and scalar multiplication. These operations must satisfy eight axioms: closure under addition and scalar multiplication, commutativity and associativity of addition, the existence of an additive identity (zero vector) and additive inverses, distributivity of scalar multiplication over vector addition and field addition, associativity of scalar multiplication, and the existence of a multiplicative identity. In robotics, we primarily work with the real number field $\mathbb{F} = \mathbb{R}$, giving us the real vector spaces $\mathbb{R}^n$.

The standard $n$-dimensional real vector space $\mathbb{R}^n$ consists of all ordered $n$-tuples of real numbers, with component-wise addition and scalar multiplication. Vectors in $\mathbb{R}^n$ represent quantities with both magnitude and direction, such as the position of a robot's end-effector in Cartesian space, the joint velocities of a manipulator arm, or the forces and torques acting on a mechanism. The dot product (inner product) between two vectors $\mathbf{a}, \mathbf{b} \in \mathbb{R}^n$ is defined as:

$$\mathbf{a} \cdot \mathbf{b} = \sum_{i=1}^{n} a_i b_i = \mathbf{a}^\top \mathbf{b}$$

This inner product induces the Euclidean norm (or $L_2$ norm) $\|\mathbf{a}\| = \sqrt{\mathbf{a} \cdot \mathbf{a}} = \sqrt{\sum_{i=1}^{n} a_i^2}$, which measures the length or magnitude of a vector. The angle $\theta$ between two non-zero vectors satisfies $\cos\theta = \frac{\mathbf{a} \cdot \mathbf{b}}{\|\mathbf{a}\|\|\mathbf{b}\|}$. When $\mathbf{a} \cdot \mathbf{b} = 0$, the vectors are orthogonal, a relationship central to many decompositions and projections in robotics.

A linear transformation $T: V \to W$ between vector spaces satisfies $T(\mathbf{u} + \mathbf{v}) = T(\mathbf{u}) + T(\mathbf{v})$ and $T(\alpha \mathbf{v}) = \alpha T(\mathbf{v})$ for all $\mathbf{u}, \mathbf{v} \in V$ and $\alpha \in \mathbb{F}$. When we fix bases for $V$ and $W$, any linear transformation is represented by a matrix $\mathbf{A} \in \mathbb{F}^{m \times n}$, where $m = \dim(W)$ and $n = \dim(V)$. The action of the transformation on a vector $\mathbf{v} \in V$ becomes matrix-vector multiplication $\mathbf{Av} \in W$.

In robotics, matrices representing linear transformations appear in numerous contexts. The Jacobian matrix of a robotic manipulator relates joint velocities to end-effector velocities through the relationship $\dot{\mathbf{p}} = \mathbf{J}(\mathbf{q})\dot{\mathbf{q}}$, where $\mathbf{q}$ is the joint configuration and $\mathbf{p}$ is the end-effector pose. The Jacobian is fundamental to velocity-level control, singularity analysis, and redundancy resolution. Similarly, the mass matrix in robot dynamics relates joint accelerations to joint torques through $\mathbf{M}(\mathbf{q})\ddot{\mathbf{q}} + \mathbf{c}(\mathbf{q}, \dot{\mathbf{q}}) + \mathbf{g}(\mathbf{q}) = \boldsymbol{\tau}$, where the mass matrix encodes the inertial properties of the manipulator.

Matrix Decompositions

Matrix decompositions factor a matrix into products of matrices with special structures, revealing properties obscured in the original representation and enabling efficient numerical computation. Several decompositions are particularly important in robotics.

The LU decomposition expresses a matrix $\mathbf{A} \in \mathbb{R}^{n \times n}$ as $\mathbf{A} = \mathbf{L}\mathbf{U}$, where $\mathbf{L}$ is lower triangular and $\mathbf{U}$ is upper triangular. When partial pivoting is incorporated for numerical stability, we obtain the LUP decomposition $\mathbf{P}\mathbf{A} = \mathbf{L}\mathbf{U}$, where $\mathbf{P}$ is a permutation matrix. The LU decomposition enables efficient solution of linear systems $\mathbf{Ax} = \mathbf{b}$ through forward and backward substitution, requiring approximately $\frac{2}{3}n^3$ flops instead of the $\frac{4}{3}n^3$ flops needed for Gaussian elimination with back-substitution when the decomposition is precomputed.

The Cholesky decomposition applies to symmetric positive definite (SPD) matrices and takes the form $\mathbf{A} = \mathbf{L}\mathbf{L}^\top$, where $\mathbf{L}$ is lower triangular with positive diagonal entries. This decomposition requires roughly half the computations of the LU decomposition and provides the matrix square root, which appears in uncertainty propagation and sampling-based algorithms. The mass matrix $\mathbf{M}$ in robot dynamics is SPD (assuming positive inertia), making the Cholesky decomposition valuable for dynamics computation and control.

The eigenvalue decomposition (EVD) applies to diagonalizable matrices and expresses $\mathbf{A} = \mathbf{V}\mathbf{\Lambda}\mathbf{V}^{-1}$, where $\mathbf{\Lambda} = \text{diag}(\lambda_1, \lambda_2, \ldots, \lambda_n)$ contains the eigenvalues and the columns of $\mathbf{V}$ are the corresponding eigenvectors. For symmetric matrices, which are always diagonalizable with real eigenvalues, the EVD simplifies to $\mathbf{A} = \mathbf{Q}\mathbf{\Lambda}\mathbf{Q}^\top$, where $\mathbf{Q}$ is orthogonal ($\mathbf{Q}^\top\mathbf{Q} = \mathbf{I}$). The eigenvalues reveal intrinsic properties of the matrix: their magnitudes indicate amplification factors, their signs determine stability in dynamical systems, and their condition number $\kappa(\mathbf{A}) = |\lambda_{\max}|/|\lambda_{\min}|$ quantifies sensitivity to numerical perturbations.

The singular value decomposition (SVD) generalizes the EVD to arbitrary matrices $\mathbf{A} \in \mathbb{R}^{m \times n}$:

$$\mathbf{A} = \mathbf{U}\mathbf{\Sigma}\mathbf{V}^\top$$

where $\mathbf{U} \in \mathbb{R}^{m \times m}$ and $