Scan Matching 2D Toolbox - DEPRECATED

The Scan Matching 2D Toolbox is a collection of 2D ScanMatching functions for range finder sensors. It contains a set of 12 algorithms plus a simple random world generator. It lets the user test algorithms with different parameters and different world and it is important to understand how ScanMatching and SLAM works. The user will be able to work and modify an open source MATLAB code for its own purposes. This document explains how the toolbox can be executed and defines the internal structure so that the user can modify the code.

Notes: This investigation was performed in 2011 and it has not been kept up to date. Most of past years investigation relies on 3D scan matching thus this algorithms are still somewhat actual. The purpose of the package is purely educational. The code is not optimized nor efficient.

Simple Execution

To execute the file, simply include the root directory in your MATLAB path, then, type (or run)

slam

At this point a random map will be generated and displayed on the screen and the user will have to draw the robot trajectory with the mouse then press Enter to start the computation. The default parameters will generate a random structured world and will correct the trajectory using an EKF based SLAM and ICP.

ICP Correction Example

Scan Matchers configuration

The following are the sensitive parameters for each aglorithm. The table shows the default and recommended parameters which are valid for most use cases. Playing with these parameters could deliver better or worse results.

Tables	Angular Threshold	Radial Threshold	Grid Resolution	Min. Iterations
Parameter	Opt.scanmatcher.Br(1)	Opt.scanmatcher.Br(2)	Opt.map.resolution	Opt.scanmatcher.niterconv
ICP	-	0.3 + σ	-	3
IDC	0.1 + σ	0.3 + σ	-	3
MBICP	-	0.5 + σ	-	3
PIC	-	-	-	3
NDT	-	-	3	3
LF-SOG	-	0.3 + σ	-	3
GENETIC	0.2 + σ/2	0.5 + σ/2	0.1	5
MONTECARLO	0.2 + σ/2	0.7+σ	0.1	4
GMAPPING	0.2 + σ	2+2σ^2	0.1	5
ANGLEHISTOGRAM	-	-	1	-
FOURIER	-	4+2σ^2	0.5	-
HOUGH	-	1+σ	4	-

Other Configurations

In order to change the following parameters. Modify the file Etc/cfg.m

input_traj = 1

Specify whether the user is going to select the robot trajectory o r using a default one.

world = wall_generator(0.1,1)

The world generation changing the object wall_generator. The first parameter controls how many ellipses will define the world. More ellipses will generate a less shaped world. If the first parameter is 0, then, the world will be unstructured, ie. It will simulates outdoor natural enviroments (caves, rocks). The second parametrs control whether the user has to insert the trajectory. Do not change, use the previous variable to work modify this.

DEBUG

This struct contains an entry for each algorithm and control whether to plot the algorithm as it is functioning.

Opt.map

Controls the mapping function for the algorithms. Used by some of them (please refer to the paper). The parameter Opt.map.resolution controls the resolution of the working space for some of the algorithm and must be changed according to the optimal parameters (listed below). Vary this parameter to test how the algorithms behave.

Opt.plot

Controls the plotting. Leave unchanged except

Opt.plot.robot_scale

increase if the robot figure is not visible or too big.

Opt.filter

Opt.filter.usefilter says whether to use a stochastic filter or not. Opt.filter.type tells which filter. (only ekf 2 d availabe).

Opt.scanmatcher

Controls most of the scanmatching parameters.

usematcher

Do scan matching?

motionamount

Travel X distance before doing another scanmatching (m)

handle

Tells which algorithm to use for scan matching. Upon changing this, remember to change the parameters as in the scan matcher optionsl list below.

associate

Specify which association method to use. Only if 'handle' is set to 'icpbSM'.

register

Specify which registration method to use. Only if 'handle' is set to 'icpbSM'.

rejection

Only value @x84. Uses x84 rjection rule during association. Only for ICP based methods

projfilter

Use projection filter to remove points not visible. Available for all

convalue

Convergence accuracy. The higher is this value the slower will take to converge. (might never converge if too high, but will stop when the max number is reached)

Opt.scan

Controls the scan forming. The only availabel options tells the maximum number of points before completing a full scan. Generally leave unchanged

Opt.error

Tells whether to display the SLAM result