Reviewed Paper Volume 2 Issue 4 December 2014 International Journal of Informative & Futuristic Research ISSN (Online): 2347-1697 A Survey On Simultaneous Localization And Mapping Paper ID IJIFR/ V2/ E4/ 029 Page No. 1056-1062 Research Area Mobile Robotics Key Words Autonomous Mobile Robots, Sensors, Simultaneous Localization And Mapping, Markov, Occupancy Grids, Adarsh M Davanageri 1 M.Tech Student, Department Of Computer Science & Engineering B. M. S. Institute Of Technology & Management, Bengaluru -Karnataka Abstract Simultaneous localization and mapping, usually called as SLAM is the technique used for construction of maps in new environment which digital machines experience. The SLAM technique is also used for updating the maps which already exist. When a digital machine, like a mobile robot visits the location for which map already exists, the new changes in environment are recognized and the changes are given as the feedback to system which is controlling the machine. The control system takes the feedback values for the input and after carrying out the verifications and mathematical calculations, decides whether to treat the feedback input is a advice to update the map's content or as an error. Simultaneous localization and mapping technique comes with many steps which have to be carried out accordingly. The main job of SLAM is to check the position of the robot every now and then, with aid of mobile robot's environment where it is standing at any point of time. By following this step the robot can have the information where it is in the world. 1. Introduction Robotics is the sub-branch of mechanical engineering; much treated under Mechatronics, electrical engineering and computer science engineering and is taken as the research area of interest. Mechanical engineers deal with the hardware and locomotive parts if any, with help of electrical engineers for the integration of boards and logical circuits. On the other hand, electrical engineers work hard with the building of circuits and other logical boards in which the logics, in the form of programs in different programming languages are written. Whole together all the three share the equal parts of building it. In recent decade, a lot of research in the field of robotics had happen. Robotics now is just not an area of fantasy to the researchers and scientists as the technology and concepts are being implemented in many of the fields. Its main advantage is taken by the medical equipment s. Different equipment s are built www.ijifr.com Copyright IJIFR 2014 1056
and also being prepared for some of the unique operations where the human hand cannot reach in the human body and it is taken care that the prepared machine is reliable to an extent that the operations made through it give good results with very less or no side effects. Many of the industries are also using robots for their production from which they can increase the production number in less time and more efficiency with no human errors. The mobile robots are those intelligent systems which are locomotive in nature. They perform different tasks while they are moving around. Mobile robotics is the field where these kind of machines come under. Mobile robots encounter many problems. When in outdoor environment they are made to sense many things at every single point of time. Popular techniques which are used in indoor environment, even with 100% accuracy, are normally preferred very less as the external environment, for a mobile robot, seems very dynamic which keeps changing in extreme extent. The problem I am going to present in this paper is one of those problems which are needed too basically by an autonomous mobile robot, the problem of simultaneous localization and mapping. Every autonomous machine should have least idea of the position where it is at any point of time and the environment around it. Without considering these things it is very hard to apply any further calculations or techniques in a dynamic environment. 2. Survey On Literature The word robot was first coined by Karel Capek and his brother in the year 1921 where they used the word for naming a character in their play Rossum's Universal Robots, a science fiction. In the play, a factory was to produce artificial people called robotics. Further, the innovation in the field of robots leads to the development of mobile robots, which are the automatic intelligent machines capable of locomotion. During the Second World War, in the years 1939-1945, the robots were used as the flying bombs, automated-pilot less aviation and detonation. As the robots become mobile and started moving around, many problems arose. Those problems include encountering an obstacle which was the primary to be solved. What decision to be taken when a mobile robot encounters an obstacle? The other problem which I am going to discuss here in this paper is "the problem of localization and mapping". Whenever the mobile robot is exposed into an environment for which a priori map data is not available, what must be the first thing must it do? How a mobile robot finds an answer to where it is standing right now? To find answers to the above queries of a mobile robot, there are many algorithms available. The way each algorithm helps depends on the kind of environment the mobile robot experiences at a given point of time. Algorithm types do vary from indoor to outdoor. Indoor environments are hardly meant for changing and some basic logics for movement can be implemented. But the scenario in the outdoor environment is totally different from indoor. The environment outside the home or office is not of that safe to an autonomous mobile robot. The landmarks, roads, vehicles and many things are subjected to change and keeps on changing over time. In such cases, a mobile robot cannot have full belief over the previous knowledge about the environment it has. So to handle the situation the use of robust algorithms is made. This report will be giving explanation about some of those good algorithms which can be used or already being used. A mobile robot can use various sensors available to localize in a known environment. Different sensors used are as in the below table. Each sensor has its own sensing function located at several parts on body of mobile robot. 1057
Table 2.1: General Classification Of Sensors Basic Classification Sensor system 1 Tactile sensors: used in detection of obstacles Bumpers 2 Motor or Wheel Sensors: speed and position Brush encoders, Optical encoders, Inductive encoders 3 Heading Sensors: pose of the robot Compass, Gyroscope 4 Ground-Based beacons: fix reference localization GPS, Reflective beacons 5 Active Ranging: time-of-flight, reflectivity Ultrasonic Sensor, Laser Rangefinder, 6 Motion or Speed sensors: to compute speed relative to a fixed reference Doppler 7 Vision Based: object recognition CMOS used in cameras, Object Tracking Packages 3. Localization A mobile robot must be able to build a map or use a prior map and then localize itself in it. If a existing map is available, the mobile robot should use it to localize itself in the environment by extracting its features and comparing with the map. If a priori map is not available, the mobile robot should make proper decisions to build a map and try to localize itself in newly built map. There are many techniques used for localizing. Localization can be carried by the information provided by the map and the sensors. The data from both the source are combined together. The control system inside the machine decides the location and the pose of the robot. Localization is also be made possible by taking the reference of active beacons like radio waves and the GPS, passive beacons like visual sensor data fusion. The autonomous machine gets the information about any landmarks or the saved plots of the environment from the given maps. When it does sense the environment and the matches the sensor data into the given apriori map, it will be confident enough to decide where it is standing in the environment.-s are done by Gamini[1] by using many of the key algorithms on a standard road vehicle The changes in the environment makes the machine hard to take reference of the landmarks and probably end up in locked state. So the robot senses the nearby area as particle. By using the particle sensors it treat every nearby field as a particle component. 3.1 Monte Carlo Localization Method Monte Carlo technique of localization uses the algorithm of particle filter localization where the particle sensors are used. The robot guesses the environment. Based on its belief by various random samples of guesses the particles are formed. The more visited location or the more believed will have the more amount of particle density. In the other hand less guessed or less visited nodes of locations Figure 3.1: Localizing Using Particles 1058
will have less density of particles. Position of the robot is calculated by the conditional probability of the distribution of the particles. The environment is divided as the particles or chunks of particles. Given the map and left in the environment, the robot starts guessing and distributes the particles. By this sampling [2] based representation we obtain a localization method which is very effective. At each point of time and space it does localize itself into the map as well as locally in the indoor environment. Larger the density of the particles, the possibility of existence of a guessed locality will be more. Hence, the particles are much connected and collected at the known locations or at the spots the guessing was made very good. As the machine moves, it keeps on making the guesses and distributing the particles. The less connected, less collected space of particles in the map of an environment that the chances are localization at the spot is very less as compared with the map. This algorithm is not much suited in the external environments. It mainly uses the light or the laser sensors. So its results are good in the closed environment. In outdoor the intensity if the light and noise will be very huge compared to indoor. In order to overcome this disadvantage, the Markov localization technique can be used. 3.2 Markov Localization Markov localization is best fit in the dynamic environments. Unlike Monte Carlo localization, Markov localization do not depend on the estimation of previous results. Markov localization predicts the results based on the present believes and sensor data. Given a map, this technique of localization takes the present observations rather than taking the information and data from the history. It is a technique [3] by which we can achieve estimation of the position of the robot in the environment. Also, each time a pose candidate is used to update the hypothesis of the pose and at this time, there is a possibilities of occurring of association errors [4]. The possibility of localizing is calculated by the probability of encountering of a landmark or a previously guessed spot. Although the present technique is well suited in the dynamic environments, the Markov localization will treat the environment as static. It computes the environment by treating it as a set static field. It assumes the environment to be static. It called as Markov Assumption. To know this technique well an example is useful. Here the mobile robot is placed in a hallway which consists of three doors. The aim assigned to the robot is to locate itself in front of any one of the doors. Before localizing itself, the robot must move around and visit all of doors in the hallway. Figure 3.2: Illustration of Markov Localization Method 1059
Initially the robot starts its journey from the first door. At this point the belief of getting a door all through the hallway will be zero, that is, nil. As it will not know what it may experiences next the mind of robot will without any data about the presence of a door? The robot moves whole path of the hallway and experiences the presence of the three doors. So the belief of the robot about presence of doors in the hallway is distributed equally. The probability of presence of doors is equal in this step. This ends the first iteration in the process of localization. So at this stage, the robot has idea of the doors and there location of presence. Whenever it tries to search for another door in the hallway it uses this probability of appearance of the door. When the robot moves through the hallway and encounters another door, it uses the same belief and again it expects that another door will in the same distance which it covered reaching the present door from the previous. It carries the belief along with its journey in the hallway as seen in the figure. By the belief it had, the robot moves further and when it comes to the fact that there is no door and the guess was false, it returns to the location where the door is present and then compares its present belief with the beliefs of presence of doors did in the first iteration. It updates the belief that the present place where it standing has high possibility of door. Here the robot is standing at a place where both the sides of it a door is present. At this point the belief of getting a door again is high. The robot thinks that it is at any one of the two doors. When it starts moving right to the hallway towards the last door the probability starts decreasing. Also when it starts moving to its left it again loses the confidence of getting a hard probability of getting a door. To localize anywhere in the map of hallway at any door it compares its present belief with the first belief of doors' location. It reaches a point where the probability was high. And localizes itself at that point in the end. The complete procedure consists of three iterations. The control system of the robot fuses all the data and finally decides where to localize. 4. Mapping The mapping and localizing is a chicken and egg problem. At certain points it is hard to consider carrying out what first, either the localization or mapping. Concurrently building map and localizing the robot in it is the best practice being followed. Many of the techniques are based on this concurrency and all try to find a solution for the problem of simultaneous localization of mapping. The context of mapping does not limit it extent only for mapping but it is also useful in path planning for a mobile robot in the environment. Figure 4.1: Building phases of map 1060
Once the proper map is produced the mobile robot feels easy to locate itself in some part of it and localize. But building a proper map is really a hard problem. The process of building a map goes through many phases. Once the autonomous mobile robot is localized with reference to the world coordinates it should build a new map, if apriori map is not available, or should start using a map. Mapping includes simultaneously estimating the pose of the robot and the map. 4.1 Occupancy Grid Mapping As the name is, occupancy grid mapping, the world is discretized into various number of cells of equal size. Initially it is to be assumed that each grid in the set are individual. A location is to be seen whether it is empty or not. If it is empty, then it is always treated empty. But if the grid is not empty, it should be further estimated that how probable it is covered by an obstacle or occupied by an obstacle. Figure 4.2: Discretizing into grids After discretizing the world view into grids it should be sensed. Each cell will be assumed to as free or occupied. In an given environment, after dividing the layout into grids, the robot must take the sensor readings. According to the sensors' data it can be decided that which grid is occupied and which grid is not. An example for the above scenario can be retrieved as in figure Figure 4.3: Occupied Grids and Empty Grids 1061
By the above sensor observation the robot will decide where to map its path. It follows the white grids and then avoids obstacle by avoiding the grey cells. The system assigns the probability of presence of an obstacle for each cell. The probability will be set with the binary random variables 0 and 1. If the grid is occupied, its probability will be assigned as 1 and if the grid is free then the probability assigned to it will be 0. There will be no dependency between each of the grids and each grid is treated individually. Getting the sensor data and the pose of the map and the robot it is easy to localize and build a map or upgrade it. The control system inside the machine decides the location and the pose of the robot. Localization is also be made possible by taking the reference of active beacons like radio waves and the GPS, passive beacons like visual sensor data fusion. The autonomous machine gets the information about any landmarks or the saved plots of the environment from the given maps. When it does sense the environment and the matches the sensor data into the given apriori map, it will be confident enough to decide where it is standing in the environment. 5 Conclusion A survey on different techniques of simultaneous localization and mapping is done. Mobile robotics is a blooming area for research in recent decades. Many thesis and new concepts keep raising daily. This work elaborates the much information about robotics. Starting from the history of robotics, the main problems of the robotics. The problem of simultaneous and localization and mapping is discussed. Many famous techniques which are implemented and their way of working have been described. By making use of the old and traditional techniques of mapping and localizing and adding some of the changes, major changes can be seen. Many techniques for mapping in the dynamic environments and many algorithms for mapping are available. Here they are made to understand well. References [1] A Solution to the Simultaneous Localization and Map Building (SLAM) Problem, M. W. M. Gamini Dissanayake, Member, IEEE, Paul Newman, Member, IEEE, Steven Clark, Hugh F. Durrant-Whyte, Member, IEEE, and M. Csorba, IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 17, NO. 3, JUNE 2001. [2] Monte Carlo Localization formobile Robots Frank Dellaerty Dieter Foxy Wolfram Burgardz Sebastian Thruny Computer Science Department, Carnegie Mellon University, Pittsburgh PA 15213 Institute of Computer Science III, University of Bonn, D-53117 Bonn. [3] Markov Localization for Mobile Robots in Dynamic Environments, Dieter Fox,Wolfram Burgard, Sebastian Thrun, Journal of Artificial Intelligence Research 11 (1999) 391-427 [4] Active Global Localization for a Mobile Robot Using Multiple Hypothesis Tracking, Patric Jensfelt and Steen Kristensen, IEEE Transactions On Robotics And Automation, Vol. 17, No. 5, October 2001 [5] Learning Occupancy Grid Maps With Forward Sensor Models Sebastian Thrun School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 [6] Computational Principles of Mobile Robotics, 2nd edition, Gregory Dudek, Michael Jenkin BIOGRAPHY Adarsh M Davanageri, a post-graduation (M.Tech) student, Dept. Of computer science & engineering, BMS Institute of Technology & Management, Avalahalli, Bengaluru, Karnataka. Have completed bachelor of engineering in computer science & engineering from S.J.M. Institute of Technology, Chitradurga, Karnataka. Areas of interest are mobile robotics, networking, virtualization. Volunteer member, Computer Society Of India. DOB: 9 th April 1992. 1062