• First assumption is that the imageSize is 640x480=307200, with corresponding BirdEyeSize of 500x500=250000.
• Get it working after changing the perspective. Only the scale 0.185 is too small for 480x360. Shall look up in the log tomorrow.
  

12 July 2008

• Excluded and Added Include-files from project. Redefined location of Qt (D:\Programs\qt-win3.1.4)
• Some Qt-parts are missing (QGraphicsItem). Downloaded binary from http://robotics.jacobs-university.de/datasets/MapEvaluation/
• Tested LoadDataSync, and committed version 1339

11 July 2008

• Added protection otherName to LoadDataSync
• Couldn't test, because no network and no battery
• Tried to build GeoTiffviewer from source on my memory-stick. Build project-file with qmake -t vcapp.

10 July 2008

• With a small additional waiting-loop, VBtest-listen and VBtest-connect work correctly with WSS0.4 (rc-2).
• Connected UsarClient to VBtest-connect. Nothing seems to be received. Adding some Writelines showed that the buffer is full. UsarClient interprets the first 5 bytes as header. The first byte is the protocol, the second is the length of the message. Because the messageLength is longer than the bufferLength, nothing is read (yet).
• Added the function SendText (including header) to Vbtest. Header is OK, but had to remove the command-part. UsarClient and VBtest-connect exchanged the first message: [CommActor]: Hercules received text 'This is a test! 0'
• Combined UsarCommander with UsarClient. UsarClient receives Pings, UsarCommander receives PoseUpdates. Strange enough, this updates are not drawn on the screen.
• Repeated test with TeleOperation instead of AutonomousExploration.
• Added a SyncStartMessage in CommActor.OnPoseUpdateReceived, but that blows up the communication.
• Added a timer on the SyncStarts (10 seconds). The result is better (although not perfect). After a while nothing happens any more (robot doesn't reacts on commands, no updates, although everything else seems to be OK.
• Changed the timer into a counter (every 10 pose updates a synch). Works reasonable, but pose-updates need priority above patch-updates. Operator sends own patches back, which can be prevented with the named patches of Julian.

9 July 2008

• Have a partial working version (committed as version 1336):
  • Two connections are made Agent to Operator and Operator to Agent.
  • Both the Agent and the Operator are sending broadcasts over this connection, but this information is not received.
  • Probably a second conversation has to be started, to receive information
• Most important changes:
  • CommActor.vb: implemented NotifyDNSReceived (created TcpConnection, Connect and OpenConversation). Changed Key of ConnectedTo-dictonary.
  • CommAgent.vb: some extra protections, no change in functionallity
  • Removed some writelines in FrontierExploration, Manifold and UsarAgent.vb
  • ICommDevice.vb: three new function (Host, Port, OpenConversation(toRobot))
  • ICommOwner.vb: modified function NotifyConversationStarted(toRobot)
  • WssConversation.vb: three new functions: ConnectedToWss, ConnectToWss, DisconnectFromWss
  • WssDevice.vb: new function WssConnection. Modified RegisterAgent, RequestDNS, RequestSignalStrength and OpenConversation. Increased period of SignalStrengthRequests to 10 seconds.
  • WssListener.vb: added writeline
  • TcpConnection.vb: added function GetClient
  • TeamController.vb: Changed TeamNames back to capitals.
• RequestConversation only from the master will create only one connection (and hopefully conversation at both sides). That is correct, but still the Broadcast messages from the UsarClient do not reach UsarCommander.
• Coupled UsarCommander with VBtest-listen. The ping-messages are received by VBtest-listen.

8 July 2008

• Changed the naming back to capitals (TeamController.vb and commandline from Debugger).
• Strange enough, the search for Hercules in NotifyAgentConnected still fails (although the list consists of two names "Operator" and "Hercules").
• Goes wrong with ConnectedTo. ConnectedTo is the only Dictonary which is search with ContainsValue. All other use ContainsKey. Changed the dictonary to from (ConversationID, agentName) to (agentName, ConversationID)

7 July 2008

• Restarted with communication-project. First attempt ended with SocketException at UsarClient:WssListener:StartListening (Only one usage of each socket adress is normally permitted). The UsarCommander was stopped at the breakpoint WssDevice:RequestDNS. The team-configuration consisted of an Operator and Hercules. Previously I used Hercules and Achilles. The WSS-console showed that Operator and Hercules both used port 8001, because Operator was Agent 1 instead of 0.
• Updated the agent-configuration. Now a connection from hercules to Operator is established (WSS outgoing port 1730). Only, the WSS (version 0.4.rc2) crashes on a access violation. Problem was that hercules wasn't spawned correctly. The UsarSim.INIT received a response DIE. hercules continued, sent its greetings, accepted a client at port 1729 and received a REVERSEDDNSREPLY {Port 1729} {Error UnknownOrIllegalPort}. Try again with a better height.
• This time the robot is spawnded at the correct height, but still msvcp90.dll crashes. Try to create a debug version of WSS from the source.
• Converted WSS.pro with the command qmake -t vcapp (no warnings). Included at three places #ifdef QT4_2_0 in MainWindow.cpp.
• The error is RobotManagementConnection:run()line 95. A message is received, but this message is empty. Added protection in RobotManagementConnection.
• The WSS stays alive, UsarClient receives ERROR {Empty.Message} on REVERSEDDNSRPELY {Port 1855}{Error UnknownOrIllegalPort}
• Problem was an additional EOL at REVERSEDDNSREPLY. Removed the extra EOL, and now the previous reported problems occur:
  • Operator is connected to the WSS via port 1878, and hercules to the WSS via port 1880.
  • There is an established connection h->O on 1879
  • There is an established connection O->h on 1881
  • There is an established connection h->O on 1884
  • There is an established connection O->h on 1885
• The problem is located in CommActor:NotifySignalStrengthReceived. If not Me.IsConnected Then RequestDNS. Unfortunatelly, IsConnected seems to be set in RequestConversation, which is commented out.
• StoreConnected is called in NotifyGreetingReceived. A GreetingMessage is sent from NotifyConversationStarted.
• Added a NotifyConversationStarted at the end of WssDevice:OpenConversation
• Seems to be double (got the Error item with the same key already added).
• REVERSEDDNSRPELY is now correctly parsed. Unfortunatelly, NotifyGreetingsReceived is not reached, because an item with the same key is already added.

4 July 2008

• Finished the rules, no time for coding

3 July 2008

• Tried different translation, rotations and flips. Switched front, Gaussian and Obstacle filter off for clearer view.
• Stored images before and after the BirdEyeConverter. The image is not converted, probably because of negative values in one of the channels.
• Derived an independent version of the equations of perspective BirdEye view. This is done for the simple case, only a rotation around the X-axis. The idea is that the BirdEye view has a central point, along the optical axis of the camera. This optical axis makes an angle theta with the normal (in this case 22.5 deg). Through this central point a cross can be drawn. Points on the the vertical line are also along the same line in the perspective image. Only, the central point is much higher (in the image at pixel 90/360. The horizontal axis also corresponds. Currently a BirdEye width of 400 pixels is chosen, but I think that it is a better choice to make it equal to the width of the perspective image. This means that the pixels in the upper right corners are cropped. It also means that on the horizontal line there is a 1-on-1 relationship between the pixels (no loss of resolution, the line is in 'focus'). Below the central line the pixels are compressed onto a smaller baseline. This is the derivation of the smaller baseline (at the bottom of the image).
• The width Wc is equal to Xbe and Xpe. Wc can be calculated from the distance to camera Lc+ and the FOV of the camera: Wc = Lc+ sin FOV. Lc+ is the diagonal of a triangle, with the height h of the camera and the ground distance Lc as other sides: Lc+ = h / tan theta. The with Wb can also be calculated from the distance to the camera Lb+. The height h is the same, the ground distance Lb less. Lb is depended on the looking angle, which is theta-FOV/2. When the fraction Wc/Wb is calculated, the height is no longer important: Wb = Wc tan theta / tan (theta - FOV / 2. Wc was set to the image width, so Wb is known.
• Tomorrow this derivation is tested.

2 July 2008

• Graduation day of Tijn Schmits, with his thesis called Development of a Catadioptric Omnidirectional Camera for the USARSim Environment.

1 July 2008

• BirdEyeConversion is currently part of the Manifold. The Manifold is part of the GUIand not related with any agent. Every agent should have its own conversion array. For the moment, I leave it in the Manifold.
• Created the translation into MatLab, but when the following image is converted, the result is a wide image (expect a long image) and the image seemed to be flipped (in the camera image the boarder goes from lower left to upper right, in the bird-eye view from upper left to lower right):
  
• This flip is not due to x=-cy'/z' and y=cx'/z' in the code. When I change that part,the images is rotated 90 degrees clockwise.
• The flip seems to be due to the flip in OmniMapLayer:TransformBitMap. This flip should only occur when camera_angle = PI/2 (to the ceiling). The bitmap is drawn around the Patch.CenterPoint (for the PerspectiveView this should be in front of the robot (ro = tan (-camera_angle) / camera_height). The DrawingScale is the magic number (3/20).

30 June 2008

• Received the matlab code for the perspective transformation, as described in the BNAIC paper.

18 June 2008

• Made a run with OmniP2DX on the waterfront-map. Preferred to drive on the omnimap-layer, instead of the save-space layer. A lot of trouble with the obstacles (due to the backwheel ?). Checked if the low parts of the sunken sportscar were visible (and they were).
  
• Two other problems: a received a wrong formatted message from the GPSsensor (removed the GPSsensor from the ini-file) and received no victiminformation from the omnicam (wrong FOV).
• Added a victimsensor, but received no information because problems with MisPkg. The Talon has its victimsensor without parent, and that approach works.
• Build a new ImgSrv, with allows resolution of 640x480. The Hook.dll couldn't be made (no d3d8.h in PlatformSDK), but a precompiled hook was available. Sensors=(ItemClass=class'USARBot.VictSensor',ItemName="VictSensor",Parent="",Position=(X=-0.29619017,Y=-0.15428557,Z=-0.19428551),Direction=(Y=0,Z=0,X=0))
• A Yrot of 0.1396 (8 deg) works quite reasonable (just outside the Gaus). Multiple VictSensors doesn't work, because a dummy orientation of zero is assumed in Patch:ProcessVictimData. Further rotating the VictSensor will result in overestimating the distance (Location should be scalled with cos(Yrot), X is the distance))

15 June 2008

• Implemented and tested the combination:
  • WSS-0.2 (listen) - VBtest (connect) works: received 100x 'This is a test'.
  This version is also closer to the UvArescue code, because it uses the stream of a TcpConnection instead of a plain socket.
• WSS-0.2 (listen) - UvArescue (connect) partly works. WSS-0.2 gives DNSREPLY, but UvArescue expects greetings. Without greetings, a new conversation is started (again and again). Try again with UvArescue on both sides.
• Run a UsarClient -ac D:\svn\Roboresc\configs\aiboField.cfg -tc D:\svn\sroebert\Configs\nb-unreal@uva.cfg -n achilles as (listen).

13 June 2008

• Tried different combinations:
  • WSS-0.2 (listen) - WSS-0.2 (connect) works (gcc-build)
  • WSS-0.2 (listen) - WSS-0.4 (connect) works (VS-build)
  • WSS-0.4 (listen) - WSS-0.2 (connect) fails (connection refused)
  • WSS-0.2 (listen) - VBtest (connect) works (VS-build)
  • VBtest (listen) - WSS-0.2 (connect) works (restarted Robots, direct socket implementation).
• Next step is more difficult. A plain socket receives bytes. The UvArescue implementation uses a TcpMessagingConnection, which is a open stream which receives messages. Look if a can use AcceptTcpClient (with a stream) instead of AcceptSocket.
• TcpConnection is a TcpClient and a NetworkStream.
• It works, I received 1024 'aaa...'s (WSS-0.2)
• It also works with WSS-0.4 (connect). Received 100000 'This is a test'.

12 June 2008

• Solved the GDI+-bug. Seems to be related with the presence of TortoiseCVS 1.10.1 on nb-unreal. Installing TortoiseCVS 1.10.6 solved the issue.
• Steven's code works, but the OmniP2DX has no laserscanner, and receives no PoseUpdates.
• VBtest is nearly finished. One agent is listening, otherone tries to connect. The one that tries to connect gets the message that connection is actively refused (firewall is off). Same error occurs for the cpp-code provided (and build in VisualStudio). Problem with reuse of adresses?

11 June 2008

• Created a VisualBasic version of WSS-0.4-src/test.cpp. Current implementation until GETSS (works after adding Newline behind the command).

10 June 2008

• WSS Version 0.4 (RC1) is out. This version has two test-programs. Test checks one conversation. Test2 spawns 100 clients.
• Included qtwin-4.1.3 into the path, and defined QMAKESPEC=D:\Programs\qtwin-4.1.3\mkspecs\win32-msvc2005. The command qmake -t vcapp now works, although I receive several warnings about =D:\Programs\qtwin-4.1.3\features\lex.prf: Unknown replace function: escape_expand. Saw something about escape_expand in qt 4.2.0. Max claims that qt 4.3.4 is needed.
• Project seems to work. Only problem is the usage of the function sleep(seconds) (sys/time.h). VisualStudio uses Sleep(milliseconds) (windows.h); Compiles.
• Copied qt-src-4.1.3\lib from pc-unreal. Changed Linker->Input to QtGuid4.lib and QtCored4.lib. Added ws2_32.lib for winsock2.h. Added vc80.pdb from qt-src-4.1.3\src\corelib into qt-src-4.1.3\lib. Links.
• Problems because qt-src-4.1.3\lib was compiled with VC7. Copied msvc?70?.* from pc-unreal:WINNT\System32 to nb-unreal. Program now runs, but directly returns (no UsarSim world).
• Debug (with command arguments '145.18.16.32 50000 hercules achilles true') works fine (checked with breakpoint), but no information is printed to output (also when less than 6 arguments). Maybe due to missing standard library.
• The XSocket needs /MTd (multithread support) as compilation option. Seems to conflict with printf(); Maybe use Rik's trick.

3 June 2008

• Created a branch for Steven Roebert's project 'Creating a Bird-Eye View map using an Omnidirectional Camera'. The branch is a copy of the roboresc-branch (with the bird-eye view transformation at version 1133) and the edu-branch. Forgot to update the UsarLib/Dialogs and Slam directories (needed for OmniP2DX and GPS sensor).
• OmniP2Dx is now in UsarBot.ini and UsarBot\Class. As side-effect, the P2AT is now also not longer standard, but this will be corrected by Tijn (do not forget).
• OmniP2DX seems to work. CamReq should store a picture in C:\Temp, but this doesn't seem the work with this behavior or with a direct spawn. Something for Steven.

18 April 2008

• The old WSS port is 5874, the new port 50000.
• Adding connection.Connect(Me._CommDevice.Host, port) to NotifyDNSReceived creates a connection, but this connection is not stored (Me.IsConnectedTo(toRobot) fails in NotifySignalStrengthReceived), so the chain RequestDNS->NotifyDNSReceived keeps on firing. IsConnectTo(toRobot) needs WssConversation:StartConversation. StartConversation is called by Wss:OpenConversation, which needs a TcpClient (we have TcpConnection). Maybe implement a OpenConversation(ByVal socket As TcpConnection).
• WSS-0.2/test still works. WSS-0.2 listener and WSS-0.4 client werken ook.

17 April 2008

• Updated to version 1151. Should continue at CommActor::NotifyDNSReceived.
• Inspected WSS-src-0.2/test. Manually created Makefile for cygwin-g++.
• Protected signalInitiation specific to Linux with #ifdef __USE_XOPEN.
• Run test with commands './WSStest.exe IP 50000 achilles hercules false' and './WSStest.exe IP 50000 hercules achilles true'. Against expectation the listen=false robot accepts the connection, and displays the bandwidth.
• Put breakpoints at INIT, GETSS and DNS commands.
• GDI+ error for robolab.
• Run as software. INIT goes well, but DNS is not reached because this is initiated with the first GETSS exchange. WSStest does only one GETSS, and is waiting afterwards.
• Run operator in UsarCommander, and hercules as UsarClient. Now NotifyDNSReceived is reached. WSSgui reports: hercules listing on 8001; operator listing on 8000; connection source listing on port 2181 should be: 2181.
• Seems that I have to create a WSSConversation (one for each connection to another agent). The accept is part of the WssListener (one per WSSDevice).

18 February 2008

• Added a parser for the Longwood dataset. The log file format has been lost in the mists of time, but is relatively easy to understand. Unfortunatelly, the data seems to be extremely noisy, with obstacles all around the robot and strange shifts of the robot. Best result was WSM + None. Committed version 960.

Labbook 2007

• http://staff.fnwi.uva.nl/a.visser/research/roboresc/Labbook2007.html
• http://staff.fnwi.uva.nl/a.visser/research/roboresc/June2007.html
• http://staff.fnwi.uva.nl/a.visser/research/roboresc/March2007.html

Labbook 2006

• http://staff.science.uva.nl/~arnoud/research/roboresc/December2006.html

Labbook 2005

• http://staff.science.uva.nl/~arnoud/research/roboresc/December2005.html

Labbook 2004

• http://staff.science.uva.nl/~arnoud/research/roboresc/August2004.html

Formatting Help?

• Usemod Wiki Text Formatting Rules http://www.usemod.com/cgi-bin/wiki.pl?TextFormattingRules