28 September 2005

For a fair comparison, a log-file was recorded and the modifications were tested for equal circumstances.
The result was that no improvement could be detected, at the 10 static reference positions the old and new algorithm showed the same performance.
See
For the flashback and odometry modifications this can be explained from the static character of the measurements, these modifications are only active when a correct localisation is temporarily lost. Yet, analysing a logfile during a real game showed that a correct localisation (validity higher than 0.6) is never reached. Yet, this logfile only contained Percepts, no Odometry data was recorded.
The conclusion: there is still a lot of work to do in the localisation. Maybe we should look at the Newbots algorithm, which showed the best localisation during the 2005 competition.

27 September 2005

The technical report is out, it can be found at http://www.dutchaiboteam.nl/research/publications/DAT2005TechReport.pdf
Part of this report are the measurements that indicate that all three modifications in the self-localization improve the performance of the self-localisation, although the three modifications together do not give any extra bonus.
These measurements were made sequential runs, with every time slightly different circumstances.

symbol-and-basic-behavior-files.dtd contained the locations of the Team behaviors, but make was still unstable (maybe because I didn't dare to commit).
Synchronised the code-base with Jurgen, and only changed the debug-symbols and intermediate code (see https://vruchtvlees.labs.cs.uu.nl/cgi-bin/trac.cgi/file/branches/amsterdam/jurgen/DT2005/Src/Modules/BehaviorControl/GT2004BehaviorControl/Makefile?rev=238).

Tried to rename GT2004Behaviours to DT2005Behaviours, but the creation agents.xinclude-processed.xml fails.
Tried xsltproc.exe --load-trace -path 'DT2005BasicBehaviors' --xinclude ../../../Tools/Xabs2/xabsl-2.2/remove-comments.xsl agents.xml -o agents.xinclude-processed.xml, but also fails.
--load-trace indicated symbol-and-basic-behavior-files.dtd loaded <- check.

Created Bin/add_revisions.sh and commited all the revision-files to DT2005.release.
Trick was to add the arguments -not -path '*/.svn/*' -and -not -path '*/.svn' to find-command.
Ask Jurgen if he likes the script, and add it to DT2005.jurgen.

Installed DT2005.release and DT2005.sp-decis on pc-aibo.
Able to see logs from the command-line with the command 'svn --username arnoud --password XXXXXX log'.
Log of directory is superset of all changes. In principle the changelog of the directory should be the common part of each log, and each file a diff to this minimum log.

Testing if I could do a one robot demo
stick 1 contains the 30.1-june-debug version and uva-paper colortable for Red1 according to stickinfo (actual Blue4).
stick 4 contains the 7-july-debug version and uva-paper colortable for Red2, which works with tl and one cloudy window.
stick 2 contains the 17.2-variable-release and the osaka-challenge colortable for Blue3, but doesn't start up (orange ears flashing).
changed stick 4 from red2 to blue1. Localised in goal. Switched tl-lights off, started crashing against back. Blue half tl-lights on, found goal again. Blocked 1/3 off field by white wallsegments. Still able to keep, return to goal after clear, and localise from start. Saved backup at pc-ias01:C:\Arnoud\backup stedelijk * changed stick 1 from Blue4 to Red 4. Attacker was quite lost
Exchanged stick 1 to Blue1, and stick 4 to Red4. No succes, Blue1 lost its goalposition twice, Red4 attacked under 30 degrees angle.