Skip to main content

1.4 The Geometry of Reality

Stage I: The Substrate

"Coordinates are a fiction we impose on the universe. The vector is the truth."

1. Vector Spaces & Basis Vectors

A point $(3, 2)$ means nothing without a reference frame. It is an instruction: "Go 3 steps along Basis $\hat{i}$ and 2 steps along Basis $\hat{j}$."

In robotics, every joint, wheel, and camera has its own Basis Vectors. The core problem of robotics is translating "My Hand's Reality" to "The World's Reality".

The Dot Product (Projection)

$\vec{a} \cdot \vec{b} = ||a|| ||b|| \cos(\theta)$

  • Intuition: How much of $\vec{a}$ goes in the direction of $\vec{b}$?
  • Use Case: Is the robot facing the goal? (Dot product of Heading and Goal Vector).

The Cross Product (Normal)

$\vec{a} \times \vec{b} = \vec{n}$

  • Intuition: The vector perpendicular to the plane formed by $\vec{a}$ and $\vec{b}$.
  • Use Case: Finding the axis of rotation for a wheel or joint.

2. Matrix Transformations (Warping Space)

A matrix is a function. It takes a vector and transforms it. $Av = b$ Matrix $A$ warps the space where vector $v$ lives.

Rotation Matrices ($SO(3)$)

The Special Orthogonal Group in 3D. A $3x3$ matrix that rotates space without stretching it (determinant = 1). It is notoriously unstable due to Gimbal Lock (losing a degree of freedom when axes align).

Quaternions ($H$)

The fix for Gimbal Lock. A 4D number system $w + xi + yj + zk$.

  • Pros: Smooth interpolation (SLERP), no singularities.
  • Cons: Brain-meltingly non-intuitive.
  • Rule: In Robotics, we compute with Quaternions, but we debug with Euler Angles (Roll/Pitch/Yaw).

3. Eigenvalues & Eigenvectors

When a matrix transforms a vector, it usually knocks it off its span. But some vectors are stubborn. They stay on their span, only getting stretched.

$Av = \lambda v$

  • Eigenvector ($v$): The axis of rotation/scaling.
  • Eigenvalue ($\lambda$): The amount of stretch.

Robotics Use Case: Principle Component Analysis (PCA). Scan a cloud of points (LIDAR). The Eigenvectors of the covariance matrix tell you the primary axes of the object. Is it a wall? A pole? A car? The Eigenvalues tell you the variance (width/length) of the cluster.


4. Singular Value Decomposition (SVD)

$M = U \Sigma V^T$ Every matrix can be decomposed into a Rotation, a Scale, and another Rotation. This is the "fundamental theorem" of data compression and noise reduction. It allows us to find the "best fit" plane for a set of noisy sensor readings.


Deep FAQ

Q: Why $\hat{k}$ for Z? A: Convention. In ROS (REP-103), we use:

  • X: Forward (Red)
  • Y: Left (Green)
  • Z: Up (Blue) Right-Hand Rule applies everywhere.

Q: Why use Homogeneous Coordinates ($4x4$ matrix)? A: A $3x3$ matrix can rotate, but it cannot translate (move origin). By adding a 4th dimension ($w=1$), we can combine Rotation and Translation into a single matrix multiplication. This simplifies kinematic chains (Arm $\to$ Wrist $\to$ Finger) into a single chain of multiplications.