A Joint Probability Data Association Filter Algorithm for Multiple Robot Tracking Problems

Abstract

Translate Article

Estimating the position of a mobile robot in a real environment is taken into account as one of the most challenging topics in the recent literature (Fox et al., 2000). This problem can be usually explored in two ways. Firstly, a mobile robot should be able to have knowledge about its current position. The Dead-reckoning of a mobile robot may be used to update the position of the robot assuming the initial position is known. However, the encoders of a robot cannot provide precise measurements and, therefore, the position obtained by this way is not reliable. To achieve more accurate approximation of a robot’s position, measurements obtained by sensors set on a robot are used to correct the information provided by the encoders. If the mapping of a physical environment is known, the abovementioned procedure can be easily accomplished by using some well-known approaches such as Kalman filtering (Kalman & Bucy, 1961) to localize the exact position of a mobile robot (Siegwart & Nourbakhsh, 2004). However, when there is not any knowledge about the map, mapping and localization should be conducted simultaneously. The aforementioned topic is known as Simultaneous Localization and Mapping (SLAM) in the literature (Howard, 2005). In many applications, one may intend to localize other robots’ position via a reference robot. Robot soccer problems or people tracking scenarios can be fallen in the pre-mentioned category. Although this problem appears similar to the common localization algorithms, the traditional approaches can not be used because the reference robot does not access to the odometry data of each mobile robot used in localization algorithms to predict the future position of the robot. This issue may be completely perceivable in the people tracking scenario because there is not any information about the movement of people. In this case, some models should be proposed to represent the movement of each object. By defining a suitable motion model for each target and using measurements provided by a reference robot about the current position of the moving object, a linear/nonlinear state space model is constructed representing the movement of each object. The above-discussed topic can be fallen in the category of target tracking problems where the final aim is defined as tracking the position of a mobile object by a reference sensor. Because of inaccurate data obtained by sensors and uncertain motion models which may not provide reliable prediction of an object’s movement, filtering algorithms are used to extract