Simultaneous localization and mapping (SLAM) enables an agent to build a map of an unfamiliar environment while estimating where it is in that environment. This lesson introduces the problem, explains why the two tasks depend on each other, and presents the probabilistic viewpoint used to reason about uncertainty.
Localization is the process of estimating an agent's position and orientation, often called its pose. Mapping is the process of estimating the layout or features of the surrounding environment. SLAM must estimate both at once.
All flashcards
Simultaneous localization and mapping (SLAM)
Estimating an agent’s location while constructing or updating a map of an initially unknown environment.
Pose
An agent’s position and orientation.
Loop closure
Recognizing a previously visited place, allowing pose and map estimates to be revised.
Maximum a posteriori (MAP) estimate
The most likely configuration of unknown variables given data and prior information.
Topological map
A representation of which places or regions connect, without requiring precise geometry.
Quiz questions
How does an extended Kalman filter (EKF) typically handle nonlinear models in SLAM?Answer: It linearizes them around an estimate and represents uncertainty with a Gaussian.
What is covariance intersection useful for when combining estimates?Answer: Combining estimates conservatively when their statistical independence is unknown.
What distinguishes a grid map from a topological map?Answer: A grid map divides space into cells that can represent occupancy.
How can a SLAM system address moving vehicles or pedestrians that violate a static-map assumption?Answer: It can detect and track moving objects separately from the static environment.
What is a central purpose of active SLAM?Answer: Choosing future actions expected to make mapping and localization more effective.