Automation Control Systems

Introduction to Simultaneous Localization And Mapping (SLAM) for mobile robot. Navigational sensors used in SLAM: Internal, External, Range sensors, Odometry, Inertial Navigation Systems, Global Positioning System. Map processing and updating principle.
Краткое сожержание материала:

Размещено на

1. Analysis of problem and problem statement

1.1 Introduction to Simultaneous Localization And Mapping (SLAM) for mobile robot

The simultaneous localization and mapping (SLAM) problem aims to create a mobile robot that would be placed at an unknown location in an unknown environment and for the robot to incrementally build a consistent map of this environment while simultaneously determining its location within this map. The solution of the SLAM problem for mobile robots will provide means to make a robot truly autonomous.

Over the past decade the "solution" of the SLAM problem has been rather successful in the robotics community. SLAM has been formulated and solved as a theoretical problem in a number of different forms. SLAM has also been implemented in a number of different domains from indoor robots to outdoor, underwater, and airborne systems.

1.1.1 History of the SLAM Problem

The genesis of the probabilistic SLAM problem occurred at the 1986 IEEE Robotics and Automation Conference held in San Francisco, California. It was a time when probabilistic methods were only beginning to introduce into both robotics and artificial intelligence (AI). A number of researchers had been looking for application of estimation-theoretic methods to the methods of mapping and localization problems; these included Peter Cheeseman, Jim Crowley, Randal Smith, Raja Chatila, Oliver Faugeras and Hugh Durrant-Whyte. Over the course of the conference long discussions about consistent mapping took place. As the result of this conversation was recognition of consistent probabilistic mapping as a fundamental problem in robotics with major conceptual and computational issues that needed to be addressed. Over the next few years a number of key papers were produced. Work of Smith and Cheeseman [1] and Durrant-Whyte [2] established a statistical basis for describing relationships between landmarks and manipulating geometric uncertainty. A key element of this work was to show that there must be a high degree of correlation between estimates of the location of different landmarks in a map and that, indeed, these correlations would grow with successive observations.

At the same time Ayache and Faugeras [3] were undertaking early work in visual navigation, Crowley [9] and Chatila and Laumond [6] were working in sonar-based navigation of mobile robots using Kalman filter-type algorithms.

These two strands of research had much in common and resulted soon after in the landmark paper by Smith. This paper showed that as a mobile robot moves through an unknown environment taking relative observations of landmarks, the estimates of these landmarks are all necessarily correlated with each other because of the common error in estimated vehicle location. The implication of this was profound: A consistent full solution to the combined localization and mapping problem would require a joint state composed of the vehicle pose and every landmark position, to be updated following each landmark observation. In turn, this would require the estimator to employ a huge state vector (on the order of the number of landmarks maintained in the map) with computation scaling as the square of the number of landmarks.

Crucially, this work did not look at the convergence properties of the map or its steady-state behavior. Indeed, it was widely assumed at the time that the estimated map errors would not converge and would instead exhibit a random-walk behavior with unbounded error growth. Thus, given the computational complexity of the mapping problem and without knowledge of the convergence behavior of the map, researchers instead focused on a series of approximations to the consistent mapping problem, which assumed or even forced the correlations between landmarks to be minimized or eliminated, so reducing the full filter to a series of decoupled landmark to vehicle filters. Also for these reasons theoretical work on the combined localization and mapping problem came to a temporary halt, with work often focused on either mapping or localization as separate problems.

The conceptual breakthrough came with the realization that the combined mapping and localization problem, once formulated as a single estimation problem, was actually convergent. Most importantly, it was recognized that the correlations between landmarks, which most researchers had tried to minimize, were actually the critical part of the problem, and that, on the contrary, the more these correlation grew, the batter the solution. The structure of the SLAM problem, the convergence result and the coining the acronym SLAM was first presented in a mobile robotics survey paper presented at the 1995 International Symposium on Robotics Research [4]. Csorba [10], [11] developed the essential theory on convergence and many of the initial results. Several groups already working on mapping and localization, notably at the Massachusetts Institute of Technology [7], Zaragoza [8], [5] the ACFR at the Sydney [20], and others began working on SLAM - also called concurrent mapping and localization (CML) at this time - in indoor, outdoor, and subsea environments.

1.1.2 Spheres of application

SLAM applications now exist in a wide variety of domains. They include indoor, outdoor, aerial, and subsea. There are different sensing modalities such as bearing only and range only.

With SLAM techniques there are more than few possible application areas, e.g., rescue robots and service robots. One can imagine intelligent autonomous robots, which can map a destroyed environment, which is too dangerous for humans to explore, e.g., buildings after an earthquake. With the provided map, a human rescuer could locate possible victims and paths to the victims. On the other hand, there are more commercial applications like cleaning or security robots that need map of an office environment. The work on SLAM is interdisciplinary and integrates the best techniques from computer vision, statistics and informatics. To achieve the goal of localization and mapping, the common approach is to equip a mobile robot system with a wide range of sensors, e.g., laser scanners, odometry, vision, sonar or RFID-receivers. With a large array of sensors, a mobile robot can map the environment during indoor or outdoor exploration. If the robot has only access to a vision unit and odometry, however the SLAM problem becomes more challenging. In the literature this class of problem is called vision-only SLAM (VSLAM).

SLAM is an essential capability for mobile robots traveling in unknown environments where globally accurate position data (e.g. GPS) is not available. In particular, mobile robots have shown significant promise for remote exploration, going places that are too distant [25], too dangerous [26], or simply too costly to allow human access. If robots are to operate autonomously in extreme environments undersea, underground, and on the surfaces of other planets, they must be capable of building maps and navigating reliably according to these maps.

Even in benign environments, such as the interiors of buildings, accurate, prior maps are often difficult to acquire. The capability to map an unknown environment allows a robot to be deployed with minimal infrastructure. This is especially important if the environment changes over time.

The maps produced by SLAM algorithms typically serve as the basis for motion planning and exploration. However, the maps often have value in their own right. In July of 2002, nine miners in the Quecreek Mine in Sommerset, Pennsylvania were trapped underground for three and a half days after accidentally drilling into a nearby abandoned mine. A subsequent investigation attributed the cause of the accident to inaccurate maps [24]. Since the accident, mobile robots and SLAM have been investigated as possible technologies for acquiring accurate maps of abandoned mines. One such robot, shown in Figure 1.1(b), is capable of building 3D reconstructions of the interior of abandoned mines using SLAM technology [26].

1.1.3 Navigational sensors used in SLAM

Visual sensing is the most information-rich modality for navigation in everyday environments. Recent advances in simultaneous localization and map building for mobile robots have been made using sonar and laser range sensing to build maps in 2D and have been largely overlooked in the vision literature.

Given that mobile robots are required to operate in largely unstructured environments and are subject to uncertainty in their motion, they must have means of sensing the environment. As the vehicle moves, it must be able to anticipate and respond to dynamic changes in its environment. Sensing can also aid in navigation as it allows the vehicle to estimate its motion and to identify the state of the environment. Sensors play an important role in navigation and planning as they allow the vehicle to track recognizable features of the environment and to detect and respond to obstacles in its path.

Internal sensors

Internal sensors are those that provide information about the robot by monitoring its motion. Robot kinematic and dynamic equations can then be used in conjunction with this information to obtain an estimate of the position of the robot.

Odometry [12]

Wheel encoders can be mounted to track the rotation and velocity of the wheels of the vehicle. Shaft encoders can also be fixed to steering mechanisms to measure the steer angle of the wheels. These measured values can be integrate...