Simultaneous mapping and localization in mobile robotics

There is a very wide range of navigation methods for mobile robotics. The simplest are reactive methods that make a direct link between current perception and action, allowing the robot to move randomly or to follow a target; but to accomplish a complex task, it is often necessary to know the robot's position in its environment and to have a map that allows the robot to plan its movements to reach a precise goal, avoiding known obstacles. A robot vacuum cleaner, for example, may work very well with a reactive strategy of random navigation, but if we imagine a service robot that can bring objects to a disabled person, precise movement planning and localization capabilities are required.

When you want to navigate using a map, you need to solve two problems: the mapping problem to create the map, and the localization problem to estimate the robot's position. For applications that can afford a fairly cumbersome set-up, mapping can be simplified by a human operator, or by modifying the environment by adding beacons, for example. However, for very widespread applications, typically personal service robots, it is desirable for the robot to be able to produce its map autonomously, without adapting the environment or requiring any special knowledge on the part of the user. This is also the case for applications where the human being cannot access the environment, such as in a military or natural disaster context. In these cases, mapping and localization problems are interdependent and need to be solved together, giving rise to a very active research field since the 1990s: simultaneous localization and mapping (SLAM).

SLAM already has consumer applications, notably with vacuum robots, some of whose models use cameras or laser rangefinders to build a map, locate themselves and plan their movements. However, this remains a very active area of research, and constant progress is being made to enable more accurate maps to be provided, containing semantic information for example, or to operate in larger, more complex environments or over longer periods of time. For the purposes of this article, we will focus on service robotics applications in indoor environments, even if most of the concepts and techniques can be used directly or adapted to outdoor environments or other types of application such as intelligent vehicles. In these contexts, however, it is generally useful to implement a satellite-based localization system (GNSS: Global Navigation Satellite System) which, while not accurate enough to directly construct a map of the environment, nevertheless provides an absolute position estimate and limits long-term localization drift.

First, we'll present the sensors and the main types of cards used for SLAM. We will then describe the main classes of algorithms,...