Index of /~andrea/pub/research/2008-icra-calibration

 Name                    Last modified      Size  Description
 Parent Directory                             -   
 logs.zip                09-Sep-2007 12:18   75M  
 logs/                   15-Sep-2007 19:24    -   
 matlab.zip              09-Sep-2007 12:19   11K  
 matlab/                 15-Sep-2007 19:24    -   

Calibration logs and source code

These are the log files and the code used for the paper:

A. Censi, L. Marchionni, G. Oriolo “Simultaneous maximum-likelihood calibration of robot and sensor parameters”

The paper is available at http://purl.org/censi/2007/calib

Table of contents:

Log description

We kept all logs we generated; however, some of them are unusable because of bugs or other accidents (dieing batteries, etc.).

In general:

• From k01 to k05 we were still developing the software: odometry has spikes in the millions, etc. We include them here for sentimental purposes.
• From k06 to k21, we were still using the old Hokuyo firmware with which we could not synchronize the Hokuyo timestamp: the timestamp in the laser log is gettimeofday() when the laser was received. These are used in the paper.
• From k22 to k30: good logs for robot “number 2”. Finally, laser and odometry are synchronized. MUCH better results. These are used in the paper.
• From k31 to k35: good logs for another robot. We tried to strip some weight to the side to change the odometry parameters, but to no avail.
• From k40 to k45: good logs for robot “number 2”, this time with the laser mounted in a different pose. See the photos below. These are used in the paper.

Please note that:

• Some logs do not excite all parameters (ex: k06, k41)
• In some files, the robots just stays still. These are used to get the Hokuyo’s sensor model.

For each experiment, there are two logs produced: one for the laser data, and one for the odometry data.

For example, for the experiment “k42”, there are the two files:

k42l.log.bz2
k42r.data.bz2

They are both compressed with bzip2. In the scripts, a string like “k42” is called the prefix.

Running the experiments (the easy way)

We have saved all the processed logs in Matlab format, so you do not have to install any other software.

By loading the file icra08_final.mat:

>>  load icra08_final.mat

you can see the logs and the results after preprocessing and do not need to install any other software.

In icra08_final.mat, there is a variable cal holding all the logs. This does the calibration using log k45:

>> res = full_calibration(cal.logs{45})

To merge two logs together:

>> big_one = merge_experiments({cal.logs{44}, cal.logs{45}});
>> res = full_calibration(big_one)

For the iterative trimming, use:

>> calib_iterative(cal.logs{44})

Running the experiments (for though guys)

Installing the software

Prerequisites:

• The C(anonical) Scan Matcher:

  http://purl.org/censi/2007/csm

  Make sure that the executables are on your path. Moreover, make sure that also the Matlab scripts contained in the CSM package are reachable by Matlab.

  >> addpath(genpath('../../csm/'))

• CMake, for the build process of CSM:

  http://www.cmake.org

• The GNU Scientific Libraries (pre-requisite for CSM).

• Ruby, for some scripts used for handling logs.

Quick way to process a log

To process a log, use these commands:

$  cd logs
$  ./prepare_one_log_compact.sh k42

this runs scan matching, and creates two files:

k22r.m           # odometry log in Matlab format
k22l_sm_stats.m  # scan-matching results in Matlab format

Load these files in Matlab using the command:

>>  log22 = calib_load_log('k22')

Run the calibration algorithm using:

>>  full_calibration(log22) 

The following does the iterative trimming:

>>  calib_iterative(log22) 

and shows the outliers.

Running the experiments (details)

This is a complete description of what goes one when running the prepare_one_log_compact.sh log.

Odometry

1. Decompression of the logs

   $ bunzip2 k22r.bz2

2. The script prepare_robot_log.rb translates the .dat to the JSON format.

   $ ./prepare_robot_log.rb < k03r.dat > k03r.json 

3. Translate the JSON in Matlab, using the json2matlab.rb Ruby script contained in CSM.

   $ ../../csm/scripts/json2matlab.rb k03r.json

   This creates the file k03r.m.

4. Load k03r.m into Matlab.

   odo03 = k03r;

   This is a cell array (one cell per interval).

Laser

We use the sm2 program in CSM. The program has two outputs: the scan-matched log (-out) and the statistics about the matching (-out_stats).

$ sm2 -in k03l.log -out k03l_sm.log -out_stats k03l_sm_stats.json 

Convert the JSON in Matlab:

$ ~/csm/csm/scripts/json2matlab.rb k03l_sm_stats.json

Read all of this in Matlab:

$ laser03 = k03l_sm_stats;

Merging odometry and laser

Use the function

log03 = merge_logs(laser03, odo03);

Run the method using:

res = full_calibration(log03);

or, for the iterative version, use:

res = calib_iterative(log03)

Format for the laser log (.log)

The laser scans are in Carmen format.

FLASER n_rays readings ... 0 0 0  0 0 0  timestamp_sec "hokuyo" timestamp_usec 

• Distances in meters, angles in radians.
• A ray with reading = 0 is an invalid ray.
• Poses are set to zero (we don’t have odometry here).

Format for the odometry log (.dat)

Each row is in this format:

x y theta   left_ticks  left_sec left_usec right_ticks  right_sec right_usec 

• x y theta odometry as reconstructed by the robot – ignore this if you are doing calibration (mm, mm, rad).

• left_ticks, right_ticks: number of ticks from the encoders.

  For converting from ticks to radians, use the formula:

   radians =  ticks_to_rad * ticks

  where

   ticks_to_rad = 2 * pi / ( 691.2 * 4 );

  The maximum value for the ticks field is 65536, then it should go over and restart at 0.

• *_sec *_usec: timestamp in seconds and milliseconds.