Loop
Bayes filter
You might say: Kalman filter, sensor fusion
A Bayes filter predicts a belief through a motion model and then conditions it on a new measurement.
- What it does
- Fuses sensors into one state estimate with an explicit belief or covariance.
- Ask
- Which filter matches the motion, measurement, and uncertainty distributions?
Parts
- Prediction step: The prediction step propagates the prior belief through the motion model and adds process uncertainty.
- Update step: The update step combines a predicted belief with a measurement likelihood; in the Gaussian linear case, the Kalman gain sets the correction.
- Noise covariances (Q and R): Q models uncertainty in the motion or process model.
Catalogue: Filter
- Extended Kalman filter: Linearizes nonlinear motion and measurement models near the current estimate and propagates a Gaussian mean and covariance.
- Unscented Kalman filter: The unscented Kalman filter is a nonlinear Gaussian filtering method in the state-estimation family.
- Particle filter: A particle filter represents a belief with weighted samples.
- Factor graph smoother: Represents states and measurements as factors and optimizes a window or history together; this is common in SLAM back ends.
Loop: cycle
Prediction step → Update step ↺
Pattern
Sources
- Timothy D. Barfoot, State Estimation for Robotics, Chs. 6–8
The Bayes filter seeks to come up with an entire PDF to represent the likelihood of the state
Use it on your project
teardown-apply: hold "Bayes filter" (bayes-filter) from the teardown at https://teardownatlas.com/robotics/ against my project
Paste it into a coding agent that has the teardown skills.
An AI agent wrote this map with the teardown skill. It can contain errors: check the sources before you rely on a part. About and legal