Map Building and SLAM Algorithms

The concept of autonomy of mobile robots encompasses many areas of knowledge, methods and ultimately algorithms designed for trajectory control, obstacle avoidance, localization, map building, and so forth. Practically, the success of a path planning and navigation mission of an autonomous vehicle depends on the availability of both a sufficiently reliable estimation of the vehicle location, and an accurate representation of the navigation area. Schematically, the problem of map building consists of the following steps: (1) Sensing the environment of the vehicle at time k using on-board sensors (e.g. laser scanner, vision or sonar); (2) Representation of sensor data (e.g. featurebased or raw-data-based approaches); (3) Integration of the recently perceived observations at time k with the previously learned structure of the environment estimated at time k − 1. The simplest approach to map building relies on the vehicle location estimates provided by dead-reckoning. However, as reported in the literature [6], this approach is unreliable for long-term missions due to the time-increasing drift of those estimates (figure 1.1(a)). Consequently, a coupling arises between the map building problem and the improvement of dead-reckoning location estimates arises (figure 1.1(b)). Different approaches to the so-called Simultaneous Localization And Mapping (SLAM) problem have populated the robotics literature during the last decade. The most popular approach to SLAM dates back to the seminal work of