Summary of "Special Topics - The Kalman Filter (2 of 55) Flowchart of a Simple Example (Single Measured Value)"

High-level summary

The video explains, with a simple flowchart, how a Kalman filter works for a single measured value. It emphasizes that the filter is an iterative three-step process that repeatedly refines an estimate and its associated uncertainty, producing progressively better estimates of the true value (e.g., satellite position, aircraft tracking, temperature).

The instructor prefers the term “uncertainty” to “error,” but uses “error” because it is common in the literature.

Main ideas / concepts

Core loop: three calculations are repeated continuously:
1. Compute the Kalman gain.
2. Update (recalculate) the current estimate using the gain, the previous estimate, and the new measurement.
3. Update the uncertainty (error) for the estimate to be used in the next iteration.
The Kalman gain balances how much weight to give the previous estimate versus the new measurement based on their respective uncertainties: smaller uncertainty → more weight.
Outputs from each iteration (the updated estimate and its uncertainty) are fed back into the loop to compute the next gain and next estimate.
The filter converges quickly toward the true value even if the initial estimate is arbitrary.

Detailed step-by-step methodology (flowchart)

Initialization

Provide an initial estimate (can be arbitrary) and an initial estimate uncertainty.

For each incoming measurement:

Calculate the Kalman gain
- Inputs: previous estimate uncertainty and the measurement uncertainty.
- Purpose: determine relative weighting of prior estimate vs. new measurement.
Update the current estimate
- Inputs: previous estimate, the new measurement, and the Kalman gain.
- Action: adjust the previous estimate toward the measurement by an amount determined by the gain.
- Output: updated (current) estimate.
Update the estimate uncertainty
- Inputs: current estimate and the Kalman gain (implicitly using previous uncertainties).
- Action: compute a new uncertainty that reflects the reduced/adjusted uncertainty after incorporating the measurement.
- Output: new estimate uncertainty to be used in the next iteration.

Repeat the loop for each new data point until estimates converge or tracking ends.

Inputs and outputs (per iteration)

Required inputs:
- x_prev — previous estimate
- P_prev — previous estimate uncertainty (error)
- z — new measurement
- R — measurement uncertainty (error)
Intermediate value:
- K — Kalman gain (computed from P_prev and R)
Outputs:
- x_new — updated estimate
- P_new — updated estimate uncertainty (fed back as P_prev for next iteration)

Key properties / practical notes

The gain expresses trust: if estimate uncertainty is small compared to measurement uncertainty, the filter trusts the estimate more; if measurement uncertainty is smaller, it trusts the measurement more.
The process is iterative and self-correcting: results from one iteration become inputs for the next.
Robust to poor initial guesses: the filter converges (“zeros in”) on the true value over time.
Typical applications mentioned: satellite tracking, radar tracking of fighter jets, and sensor measurements like temperature or height.