A robot on a line with an encoder and a GPS: why each one fails alone, and how predict and correct combine them into the Kalman filter. 6 short slides, then worked examples and problems. Each slide has an everyday picture, a worked example and Python you can run right on the page.
Lesson 2Dead reckoningDead reckoning: adding up moves makes the estimate driftOpen slide →01 · Lesson 2 · Dead reckoningAdd up noisy encoder velocities and the position drifts. See why the variance grows by Q every step, with worked examples and Python you can run.
Lesson 2GPS onlyGPS only: no drift, but every reading is noisyOpen slide →02 · Lesson 2 · GPS onlyA GPS measures position directly, so it never drifts, but every reading jumps around the truth. See the measurement model, RMS error and Python to run.
Lesson 2PredictThe predict step: move the estimate, grow the varianceOpen slide →03 · Lesson 2 · PredictThe predict step moves the estimate with the encoder and grows its variance by Q: x̂⁻ = x̂ + u, P⁻ = P + Q. Worked examples and live Python.
Lesson 2CorrectThe correct step: fuse the prediction with the GPSOpen slide →04 · Lesson 2 · CorrectThe correct step fuses the prediction with the GPS using the gain K = P⁻/(P⁻ + R). The 35 → 38 worked example, practice problems and live Python.
Lesson 2The 1D filterThe 1D Kalman filter: predict, correct, repeatOpen slide →05 · Lesson 2 · The 1D Kalman filterRun the full 1D Kalman filter on a robot with an encoder and a GPS. See the steady state, and how tuning Q and R trades smoothness for lag. Live Python.
Lesson 2RecapLesson 2 recap: the 1D Kalman filterOpen slide →06 · Recap · ReadingThe 1D Kalman filter in one page: dead reckoning drift, noisy GPS, the predict step, the correct step with the gain K, the steady state and tuning Q and R.