SLAM-R Algorithm of Simultaneous Localization and Mapping Using RFID for Obstacle Location and Recognition

Introduction
For decades, research has focused on attempting to simulate common human actions like walking, running, talking and even thinking. One of the qualities of humans that has gathered the most attention from scientists is their ability to move around in different settings, making researchers focus on navigation techniques that transfer this ability to artificial entities. In 1986, Peter Cheeseman, Jim Crowley and Hugh Durrant-Whyte talked about the topic of simultaneous localization and mapping applying probability (SLAM), during the IEEE Robotics and Automation conference held in San Francisco, United States [1].

The creation of SLAM resulted in various research that tried to determine which action would be carried out first, localization or mapping [2], [3], [4], [5], [6], [7], [8]. Multiple algorithms allowing for the simultaneous navigation and localization (SLAM) of mobile robots have been developed since then, both for indoor and outdoor environments. Table 1 includes some of those algorithms.

The creation of SLAM resulted in various researchthat tried to determine which action would becarried out first, localization or mapping [2]-[8].Multiple algorithms allowing for the simultaneousnavigation and localization (SLAM) of mobilerobots have been developed since then, both forindoor and outdoor environments. Table 1 includessome of those algorithms.

A description of each algorithm included in Table 1follows. Algorithm GMapping [9] is a particle filterbased online algorithm with Rao-Blackwellizationproposing distribution of probabilities that considerthe last measure taken by the laser device, and notjust odometry. This is done by searching the regioncloser to the estimated location, defining theprobability of each landmark associated to themeasure and adding the odometry information; from this, K samples are extracted to estimate aGaussian distribution matching the mean andvariance with the estimated distribution. Theparticle’s new position is obtained from theresulting distribution. Before resampling, ameasure inversely proportional to the variance ofparticle’s estimations is calculated to assess theneed of resampling. This algorithm was tested withdata from Intel1, Freiburg2 and MIT3, with goodresults, generating maps without inconsistenciesfor each tested data, including analysis fromdifferent researchers. One of the setbacks of thealgorithm is dynamic objects, as well as objectswith complex modeling like grass, wires, etc.