What is SLAM?

Parent: SLAM index
See also: slam-hardware

Source:  Scaradozzi 2018

Process which allows a mobile robot to

• construct a map of its environment (assumed to be unknown)
• compute its location using the map simultaneously

Source:  Lamarca 2020

• Goal is to locate a sensor in an unknown map/environment, which is simultaneously being reconstructed.
• Typically used in exploratory trajectories (new or changing environments)

Source:  Wikipedia SLAM

Simultaneous localization and mapping (SLAM)

• computational problem
• construct/update a map of an unknown environment
• simultaneously keep track of an agent’s location within it

Source: SLAM for Dummies

Goal: to use the environment to update robot position

• robot odometry is often erroneous, therefore we cannot rely directly on odometry
• in this specific example, laser scans are used to correct the position of the robot in a basic EKF-SLAM implementation
  • extract features from environment
  • update position estimate based on the features
  • re-observe when the robot moves around

Source:  Cometlabs

• Getting precise measurements based on known maps –> ok
• Building a map of the environment –> ok

But solving both simultaneously is harder!

With a map (i.e. prior knowledge of environment, e.g. via storage of known landmarks in a database)

• mobile robot can perform a set of tasks like path planning
• error in estimating the state is limited
• robot can ‘reset’ localisation error by revisiting known areas

Without a map

• Dead-reckoning / odometry would quickly drift over time
• Data association becomes much more difficult
  • Unknown mapping b/w existing landmarks and observations

Source: Grisetti 2011 - Tutorial graph-based SLAM

Another way of defining SLAM: “learning maps under pose uncertainty”