Foundations · principles
Playable canvas figures
Every concept gets a hands-on canvas: drag the parameters, watch the numbers move, then read the article that treats it rigorously. Understand the phenomenon before the proof.
The first group follows the two bridges of the math track (from high school into linear algebra and geometry); the rest come from the lesson pages. Each figure names the page it belongs to.
T1 · The Matrix Is a Ledger for Rotation
drag the point on the circle and watch the four numbers of the rotation matrix update live — a matrix is not a table, it is a moving ledger
Read this one →
T2 · From Slope to Gradient: The First Line of Every Robot Learning Algorithm
drop the ball on a slope and drag the learning rate: too small crawls, too big overshoots, about 0.25 is just right
Read this one →
T3 · Vectors and forces: the physical reading of the dot product
drag the force arrow to change its angle and watch the work bar update — wrong angle, wasted effort
Read this one →
T4 · Two Rulers and a Better Ruler
drag the widths of two bell curves; the fused result is always narrower than either ruler — inverse-variance weighting, animated
Read this one →
T5 · The Ledger of One Motor Revolution: Radians, Speed, and Power
spin the wheel and watch radians, degrees, and line speed roll in real time — arc length s = rθ is the distance the wheel covered, not a formula
Read this one →
T6 · Complex multiplication is rotation: the one-dimensional version of quaternions
multiplying by a complex number on the unit circle = rotating by an angle; in 3-D this same thing is the quaternion
Read this one →
T7 · Bases and coordinates: why robots put everything in matrices
drag the basis vectors e₁ e₂ and watch the coordinates of the same vector recompute — change the basis, not the vector
Read this one →
T8 · Matrix Multiplication: Chaining Transformations, Not a Table Operation
the same input arrow, AB and BA land at different endpoints — drag the slider to change the input direction; it holds everywhere
Read this one →
T9 · The Jacobian: Rotate a Joint One Degree — How Far Does the End Effector Move?
drag the arm joints and watch |det J| collapse — the moment it stretches straight is the singular pose
Read this one →
T10 · Eigenvectors: The Directions a Matrix Dare Not Touch
the point cloud drifts gently and the principal-axis arrow follows — PCA is exactly this arrow that always points where the cloud spreads most
Read this one →
T11 · SVD and the Pseudoinverse: When There Is No Exact Solution, Find the Nearest One
a unit circle stretched into an ellipse — the singular values are the major and minor axes; no matrix escapes these three steps
Read this one →
T12 · Taylor expansion: putting a straight line to work in place of a curve
press a button to change the order and watch the dashed polynomial hug the sin curve — the approximation comes with a contract
Read this one →
Figures inside the lesson pages
Gimbal lock
Drag the pitch angle to 90° and watch two axes collapse into one - three numbers left with two degrees of freedom
Play it in the lesson →
TF chain
Drag a point and watch the base / end-effector / camera books change at once - the intuitive core of the tf tree
Play it in the lesson →
tf forward and inverse
Forward p_b=R·p_t+t, inverse p_t=Rᵀ(p_b−t): click the button to switch direction and watch how the matrix transposes
Play it in the lesson →
Bayes
Drag the prevalence and accuracy and watch how, among 500 people, only a little over ten percent of the positives are truly sick
Play it in the lesson →
Expected value
Roll the dice enough and the sample mean converges to 3.5, with jitter ∝ 1/√n
Play it in the lesson →
Convex vs non-convex
The same downhill rule: a ball in a bowl always reaches the bottom; a ball in a valley's fate depends on where it starts
Play it in the lesson →
Euler's method
Change the step size and watch the polyline chase the curve: the bigger dt, the more each step overshoots the decay
Play it in the lesson →
Aliasing
Undersample and a 3Hz sine turns into a fake frequency at the digital end
Play it in the lesson →
Low-pass filter
Drag the window size and watch spikes get smoothed away while latency grows
Play it in the lesson →
Pinhole imaging
Drag the object point and focal length: pixel = focal length × meters ÷ distance - that is exactly what calibration solves
Play it in the lesson →
Convolution
Change the kernel and watch the output: edge / blur / sharpen - what a conv layer learns is the numbers in the kernel
Play it in the lesson →
Stereo disparity
Drag depth and baseline: z = f·b/d, and at long range disparity turns into guesswork
Play it in the lesson →
SLAM loop closure
Watch the gray loop fail to return to its start while loop closure pulls the orange loop straight - drift spread flat across the whole map
Play it in the lesson →
Complementary filter
IMU is fast but drifts, camera is accurate but slow - the τ knob decides whom to trust
Play it in the lesson →
Value iteration
Watch scores propagate back from the goal; after 30 sweeps the whole corridor grows a policy
Play it in the lesson →
Diffusion denoising
Watch pure noise ground step by step into an action trajectory - multimodal, never averaged
Play it in the lesson →
VLA pipeline
See the image, hear the command, emit action chunks: everything pressed into one token stream
Play it in the lesson →
Tactile array
Drag the contact point on an 8×8 pressure heatmap and watch the centroid tell whether the grip is straight
Play it in the lesson →
Move vs copy
Does a 100MB object fly over or get copied: what std::move is for
Play it in the lesson →
Vectorization
Pure Python loop vs NumPy: a hundredfold gap that widens with data size
Play it in the lesson →
epoll
Four threads blocking vs one thread event-driven: overlapping waits double throughput
Play it in the lesson →
Latency tail
Drag the load and watch the histogram: real-time is about the worst tail, not the average
Play it in the lesson →
Topic flow
Watch messages flow along topics; when an intermediate node drops, downstream never sees an error
Play it in the lesson →
QoS queue
Publishing faster than consuming: a full queue drops messages - KEEP_LAST's nature
Play it in the lesson →
EtherCAT
Watch a frame fly through every slave: cycles in the hundreds of microseconds, jitter <1μs
Play it in the lesson →
Frequency pyramid
Sensing/control at 1kHz, perception at 30Hz: fast-slow separation is architecture rule number one
Play it in the lesson →
Branch and merge
feature branch → merge → conflict → resolve: a normal day with git
Play it in the lesson →
Flame graph
The widest block is the hotspot; cutting 30% of hotspots makes the whole thing only 1.4× faster
Play it in the lesson →
Boot chain
ROM→U-Boot→kernel→services: quantify segment by segment to cut segment by segment
Play it in the lesson →
Yocto cache
Change 1 recipe, rebuild 12 packages - the other 130 come from the sstate cache
Play it in the lesson →
Executor
Blocking in a callback = the whole team queues up; multithreading parallelizes but you must lock yourself
Play it in the lesson →
Zero-copy
A 2MB image copied three times per frame vs pointer passing: 168 vs 24 MB/s
Play it in the lesson →
RC charging
63%@1τ, 99%@5τ: drag R and C - the yardstick for everything fast and slow
Play it in the lesson →
PWM generation
Watch the counter sawtooth sweep past CCR: duty cycle and average voltage at a glance
Play it in the lesson →
FOC
Watch three-phase AC move into the rotating frame: Iq makes torque, Id only makes heat
Play it in the lesson →
Encoder AB phases
Whoever leads decides the direction: 4x counting and direction decoding for incremental encoders
Play it in the lesson →
Thermal path
Steady-state junction temp = ambient + power × thermal resistance: drag the power and watch it cross the red line
Play it in the lesson →
Gear ratio
Torque ×i, inertia ×i²: drag it to 50 and watch inertia pin the motor against friction
Play it in the lesson →
Common-mode cancellation
Both wires carry the same noise; subtract and the signal doubles while noise zeroes out
Play it in the lesson →
Compute budget
List workloads item by item, discount effective compute: the multiplication problem of domain-controller selection
Play it in the lesson →
delay vs millis
delay freezes the whole show, millis interleaves tasks: the gate from beginner to engineer
Play it in the lesson →
Wireless retransmission
Drag the distance and watch drops and retries: WiFi guarantees data, not latency
Play it in the lesson →
Priority inversion
A low-priority task holding a lock stalls the high-priority one; priority inheritance rescues it in 15ms
Play it in the lesson →
PCB impedance
Drag the dielectric thickness and watch the characteristic impedance: 50Ω is calculated, not guessed
Play it in the lesson →
Scope trigger
Scroll left, stay steady right: the trigger makes every sweep start from the same event
Play it in the lesson →
Termination reflections
An unterminated step is all reflections: at 3cm of trace, treat it as a transmission line
Play it in the lesson →
Eye diagram
Drag the jitter and watch the eye close: a checkup chart for decision margin
Play it in the lesson →
I2C waveform
START, address, ACK, STOP: understand a two-wire bus in three steps
Play it in the lesson →
USB enumeration
Six steps from plug-in to driver loaded: each failure step maps to one field symptom
Play it in the lesson →
CAN arbitration
Bit-by-bit ID comparison, dominant over recessive: the lowest ID wins the bus without loss
Play it in the lesson →
Camera bandwidth
Pixels × frame rate × bit depth × 1.25: the multiplication problem of interface selection
Play it in the lesson →
Simulation step
Too big a dt and the ball explodes: contact is a super-stiff spring
Play it in the lesson →
System identification
Drag inertia and damping onto the real-robot curve: the residual is the physics you did not model
Play it in the lesson →
HIL latency
A real controller meets a fake world: inject latency and watch when the loop falls apart
Play it in the lesson →
Annotation IoU
Drag the predicted box and watch IoU: the ruler of annotation quality
Play it in the lesson →
Bisect
16 versions, 4 steps to lock down the one that introduced the bug: automatic demo
Play it in the lesson →
Five-level light-up
Power→comms→drivers→sensors→closed loop: no green at one level, no entry to the next
Play it in the lesson →
Risk matrix
Severity × frequency sets the safety level: where functional safety starts
Play it in the lesson →
Real-time factor
Physics + sensing + rendering make one cycle: do not trust an RTF<1 simulation
Play it in the lesson →
RRT
Watch a random tree grow past obstacles: the escape hatch of high-dimensional space - crooked paths, but it finds one
Play it in the lesson →
Control loop
read→update→write: drag the frequency and see whether one tick is enough
Play it in the lesson →
Costmap
Watch a wavefront spread backward from the goal: the core of Nav2 global planning
Play it in the lesson →
Domain randomization
Randomization magnitude is an inverted U: too small overfits the sim, too big makes the policy give up
Play it in the lesson →
Soft contact
Penetration × stiffness = contact force: MuJoCo's spring world
Play it in the lesson →
Convergence compared
Drag the conditioning κ: gradient descent cannot climb, L-BFGS converges in a few dozen steps
Play it in the lesson →
Playback interpolation
Multi-topic timeline + playhead: rosbag is the black box
Play it in the lesson →
Data pipeline
Add DataLoader workers and watch the GPU go from waiting for food to eating its fill
Play it in the lesson →
Quantization
Watch weight distributions become stair-stepped: INT8 with 0.4% error at 2–3× speed
Play it in the lesson →