chlai88 / opennero Goto Github PK
View Code? Open in Web Editor NEWAutomatically exported from code.google.com/p/opennero
License: Other
Automatically exported from code.google.com/p/opennero
License: Other
Purpose of code changes on this branch:
When reviewing my code changes, please focus on:
Is the Z-Score being calculated properly
After the review, I'll merge this branch into:
/trunk
One of the reasons that Explore/Exploit isn't working correctly right now could
be that the Z-Score is broken, we should look into this and make sure that it's
working right.
Check Revision 275 for the last change to the code in
/mods/NERO/NeroEnvironment.py
Original issue reported on code.google.com by [email protected]
on 8 Jul 2010 at 8:12
What steps will reproduce the problem?
1. Try to right click very far away from the camera in NERO
2.
3.
What is the expected output? What do you see instead?
Doesn't form on the correct place
The suggested solution to this is to use a hidden plane at Z=0 for intersection
tests with the right-click menus and to ignore the landscape entirely.
Original issue reported on code.google.com by [email protected]
on 19 Sep 2010 at 7:29
Test the first-person agent in the mage.
* Start OpenNERO (trunk)
* Start the Maze mod
* Click on First Person agent
* Complete the maze
* Repeat before/after running other agents
* Repeat before/after/during closing the external plot server
Evaluation questions:
* Does the first person control make sense and work well?
* Are the instructions on the web clear?
* Are there any issues you could find when combining with other agents?
* Does the fitness plot do what you expect?
Original issue reported on code.google.com by ikarpov
on 30 Sep 2010 at 3:11
What steps will reproduce the problem?
1. Compile trunk under Mac OS X
2. Start the program and click either NERO or Battle
3. Click "Deploy" in the OpenNERO window
4. Adjust one of the sliders in the external interface
What is the expected output? What do you see instead?
On Linux, this works fine - i.e. adjusting the values
changes the fitness function as expected. On Mac OS X
(and possibly Windows as well) this does not work and
the program stops in blue screen due to an error in
the code that is talking to the external UI (error
code is Invalid file handle).
This probably has to do with how pipes work on different
Python platforms. We should really try to use something
else for communicating with the external UI, such as a
socket - that will work on all the platforms.
Original issue reported on code.google.com by ikarpov
on 13 Sep 2010 at 3:41
Purpose of code changes on this branch:
This branch in general has three sets of changes:
* merge from the creativeit branch (in particular for object manipulation)
* rtNEAT changes from the rtNEAT package comparison work
* collision detection and collision resolution code
This review is concerned with the changes implementing collision detection.
When reviewing my code changes, please focus on:
* Roomba mod
* collisions should be detected and not allow roomba agents to pass through each other
* code for detecting these collisions is in:
* SimContext::UpdateSimulation (source/game/SimContext.cpp)
* Simulation::DoCollisions (source/game/Simulation.cpp)
* SimEntity::GetCollisions (source/game/SimEntity.cpp)
* SimEntity::ResolveCollisions (source/game/SimEntity.cpp)
* SceneObject::CheckCollision (source/render/SceneObject.cpp)
Original issue reported on code.google.com by ikarpov
on 13 Jul 2010 at 4:57
What steps will reproduce the problem?
1. Start roomba
2. Run simulation until roomba agents collide
3. Some will likely overlap
What is the expected output? What do you see instead?
Roomba agents should not overlap with each other
Please use labels and text to provide additional information.
Original issue reported on code.google.com by ikarpov
on 15 Sep 2010 at 3:43
The new "steve" bots have turn animations that we should use in Maze.
Instead of running from cell to cell and turning in-place, the bot should use
the turn animation.
Original issue reported on code.google.com by ikarpov
on 29 Sep 2010 at 3:45
Purpose of code changes on this branch:
This code has three sets of changes:
- merge from the creativeit branch (in particular for object manipulation)
- rtNEAT changes from the rtNEAT package comparison work
- collision detection and collision resolution code
When reviewing my code changes, please focus on:
- rtNEAT fix from the rtNEAT package comparison (Revision #412)
- test that the new rtNEAT code compiles and runs without errors and performs no worse on the new NERO environment
After the review, I'll merge this branch into:
/trunk
Original issue reported on code.google.com by ikarpov
on 13 Jul 2010 at 3:36
What steps will reproduce the problem?
1. Run in NERO Mode
2. Click "deploy" on external UI
3.
What is the expected output? What do you see instead?
We want this to act like an actual deploy button
Please use labels and text to provide additional information.
Original issue reported on code.google.com by [email protected]
on 18 Sep 2010 at 10:03
We need some relatively high-contrast texture in the Roomba environment to
allow us to see the motion of the robots better.
Original issue reported on code.google.com by ikarpov
on 18 Jul 2010 at 7:50
We need to create a Wiki page documenting in detail the process of taking
original 3D models from Maya or 3DS studio max and including them in OpenNERO
mods.
Original issue reported on code.google.com by ikarpov
on 18 Aug 2010 at 4:00
What steps will reproduce the problem?
1. cmake -G "Visual Studio 2005 8.0"
2. build project in Debug configuration
What is the expected output? What do you see instead?
Expect to produce debug executable
See linking error because of Boost.Python automatic linking issues and/or cmake
configuration
instead
Please use labels and text to provide additional information.
Original issue reported on code.google.com by ikarpov
on 23 Oct 2008 at 4:16
What steps will reproduce the problem?
1. Open NERO mod
2. Click Deploy
3. Close menu
What is the expected output? What do you see instead?
It should not crash, or generate a menu for bringing up the external UI again.
Instead it crashes.
Please use labels and text to provide additional information.
Original issue reported on code.google.com by [email protected]
on 19 Sep 2010 at 6:45
What steps will reproduce the problem?
1. Click deploy in NERO
2. Adjust the sliders for a task (such as approach flag)
3. Wait for a while to collect some datapoints in the plot window
4. Either:
* Adjust the lifetime slider to increase the lifetime dramatically, or
* Change the fitness weights
What is the expected output? What do you see instead?
I expect the plot to be a meaningful representation of the population fitness.
What happens instead is there is a spike of huge values immediately after
change - this is because the old population average and standard deviation are
no longer valid, but are still being used to calculate the Z-score of the
currently evaluated individuals.
We should probably reset the population statistics on at least some of these
changes. Currently this prevents us from doing any kind of shaping tasks.
Original issue reported on code.google.com by ikarpov
on 23 Sep 2010 at 2:37
Steps to follow:
* Start OpenNERO Maze mod
* Run a number of A* agents and observe the animation of the new bot
* Observe the animations under different amounts of system load
* Adjust the speedup slider
Evaluation questions:
* Does the bot animation appear natural?
* Is it well synchronized with the speed of transition?
* Is the bot animation synchronized with the footsteps left behind by the model?
* Does the bot animation appear smooth?
* Does the appearance of the animation change with system load?
* What happens to the animation when the speedup slider is adjusted?
Original issue reported on code.google.com by ikarpov
on 30 Sep 2010 at 3:30
One of the functions necessary for the NERO mod to reach a usable level is the
implementation of firing mechanics. This needs to be implemented in the nero
branch, and then balanced needs to be experimented with to ensure that the
agents can learn to fire properly.
Original issue reported on code.google.com by [email protected]
on 16 Jul 2010 at 7:25
We would like to add a character to the OpenNERO resources that is better
suited for long-range combat and environment manipulation tasks, e.g. NERO mod
or Blocks World.
Dan made a working model of a bot that looks great and it should be integrated
into Maze and NERO mods.
Original issue reported on code.google.com by ikarpov
on 8 Jul 2010 at 7:37
On Windows (and other platforms that don't have wxPython installed), the new
external UI does not start and does not report a problem, causing lots of
python errors to be printed out to console instead.
We should check if wx is installed and pop up a window saying where to download
it if it is not. The popup can be done the same way as upon pressing F1 - i.e.
asking the system to open a web browser with a help page describing the issue.
Additionally, something that is going to work no matter what should also report
this problem (perhaps Irrlicht window or a blue screen).
Original issue reported on code.google.com by ikarpov
on 12 Sep 2010 at 1:29
Purpose of code changes on this branch:
trunk
When reviewing my code changes, please focus on:
r526: Ideally this should, if wx is not installed, call up a wiki page ('wx'),
and then return to hub, testing on a local machine with a non-existant package
replacing wx (skynet) does not result in appropriate behavior, but also does
not interfere with behavior. Any reason these functions aren't working properly
at that location?
After the review, I'll merge this branch into:
/trunk
Original issue reported on code.google.com by [email protected]
on 20 Sep 2010 at 3:54
Purpose of code changes on this branch:
When reviewing my code changes, please focus on:
After the review, I'll merge this branch into:
/trunk
/source/ai/rtneat/rtneat.cpp
Three is a particular section of the next_organism code which is supposed to
handle Exploit behavior -- is it working correctly?
this is under /branches/nero, but this goes back to that last version of the
code hosted on here, as far as I can tell.
Original issue reported on code.google.com by [email protected]
on 8 Jul 2010 at 8:18
http://code.google.com/p/opennero/wiki/MazeGeneratorExercise
Original issue reported on code.google.com by ikarpov
on 18 Aug 2010 at 3:50
What steps will reproduce the problem?
1. Run NERO mod or Battle mod for a while, try to move flag
What is the expected output? What do you see instead?
BSOD, id -1 not found message shown.
Please use labels and text to provide additional information.
Original issue reported on code.google.com by ikarpov
on 23 Sep 2010 at 5:23
http://code.google.com/p/opennero/wiki/MazeLearnerExercise
Original issue reported on code.google.com by ikarpov
on 18 Aug 2010 at 3:52
Steps to follow:
* Start OpenNERO Maze
* Run a number of learning and other agents
* Observe the external plot server
* Close and reopen the mod
* Close the external plot server in the middle of a run
* Use the speedup slider
Evaluation questions:
* Does the plot of fitness make sense?
* Does it appear current and accurate?
* Does closing the plot server affect execution?
Original issue reported on code.google.com by ikarpov
on 30 Sep 2010 at 3:33
What steps will reproduce the problem?
Inspect the values in agent.state.rotation.z at line 381 in Roomba module.py
What is the expected output? What do you see instead?
The expected range of values is [-180,180]. I observed several values lesser
than -500.
What version of the product are you using? On what operating system?
I'm using Linux, and checked out version 403 from /branches/merge/
Please provide any additional information below.
A quick fix for the subsequent lines of code that depend on
agent.state.rotation.z to be in the correct range could be to insert the
following code after line 381 (angle = degrees(atan2(disty, distx)) -
agent.state.rotation.z):
sign = 1
if angle < 0:
sign = -1
angle = sign * (fabs(angle) % 360)
However, this fix still doesn't address any issues that might arise in the rest
of the code due to agent.state.rotation.z being out of range.
Original issue reported on code.google.com by [email protected]
on 22 Jun 2010 at 10:10
Adam, please provide details on this bug.
Original issue reported on code.google.com by ikarpov
on 23 Sep 2010 at 2:38
When using the collision detection system to collide NERO bots with walls in
NERO or Battle, some bots go through walls because collision detection system
is not respected by the Python code keeping track of agent state information.
Most likely need to modify Python code to pay attention to CPP side updates
sooner.
Original issue reported on code.google.com by ikarpov
on 15 Sep 2010 at 3:46
What steps will reproduce the problem?
1. Right click and select Add Wall in NERO
2. Right click on the wall and select Scale
What is the expected output? What do you see instead?
The wall should be on the same level and of the same height as the walls already in the environment - it currently looks too small to be an obstacle for the bots.
When scaling, there are two problems. One is that the wall is hard to resize beyond the original bounds - it is either smaller or way too huge, and the middle range is hard to get at (if possible). This may depend on the orientation of the wall.
Original issue reported on code.google.com by ikarpov
on 23 Sep 2010 at 2:30
Testing evolution in NERO Mod to ensure evolution is happening properly.
Original issue reported on code.google.com by [email protected]
on 29 Sep 2010 at 4:51
Irrlicht GUI is limited and difficult to use, either from our C++ code or from
Python. In particular, it covers up the 3D graphics window's real-estate, looks
ugly and does not allow us to use nice native widgets such as file dialogs.
In order to address this, we would like to allow people to easily use other GUI
tools such as wxPython or Tkinter. Adam is working on providing a working
example of doing this in the NERO mod. This work is done under the
source:/branches/menu
Original issue reported on code.google.com by ikarpov
on 8 Jul 2010 at 7:29
While validating experiments against the original rtNEAT package, several minor
issues were discovered with the implementation currently in use in OpenNERO.
These issues have been addressed and need to be merged in to trunk after
testing.
Original issue reported on code.google.com by ikarpov
on 8 Jul 2010 at 8:04
OpenNERO in linux currently requires libboost1.38, which is currently not
compatible with Ubuntu Lucid Lynx.
Upgrading to libboost1.40 I believe would allow linux 10.03 (lucid lynx) to
have the same AI experience :-)
Original issue reported on code.google.com by [email protected]
on 8 Sep 2010 at 3:44
What steps will reproduce the problem?
1. Start OpenNERO
2. Load a mod with more than 20 agents on the field (such as Roomba or NERO)
3. Record the frame rate
What is the expected output? What do you see instead?
We would like to keep the frame rate above the 24 FPS range. Instead, on many
machines the rate drops to around 10.
The frame rate improves dramatically from the Debug build to the Release build.
However, in NERO and Roomba environments, the sensor calculations are not
efficient and in Roomba in particular are always done every frame for every
agent, which leads to large frame delays.
Several possible solutions:
* Split agent evaluations between frames
* Make sensor and other calculations more efficient, in particular in Roomba
* Decrease the complexity of the level geometry and the collision detection system
Original issue reported on code.google.com by ikarpov
on 18 Jul 2010 at 7:46
http://code.google.com/p/opennero/wiki/NeroMod
Original issue reported on code.google.com by ikarpov
on 18 Aug 2010 at 3:54
Currently there is no effective sensor model that works well for running
coevolution in any of the various mods. We want a sensor model that will model
the behavior of NERO 2.0
Original issue reported on code.google.com by [email protected]
on 21 Jul 2010 at 7:43
Steps to follow:
* Start OpenNERO
* Start NERO mod
* Adjust the values and the sliders on the external UI
* Look in the log to make sure the values adjusting there correspond to what you are changing
* Use the Save and Load buttons
Evaluation questions:
* Does the external UI work as expected?
* Does it affect the agent behavior?
* Are you able to adjust all the values correctly?
* What are the ranges of the values you adjusted?
Original issue reported on code.google.com by ikarpov
on 30 Sep 2010 at 3:17
http://code.google.com/p/opennero/wiki/MazeSolverExercise
Original issue reported on code.google.com by ikarpov
on 18 Aug 2010 at 3:51
Textures abruptly switch to crazy at a certain distance away from the camera
when running OpenNERO in Windows.
Original issue reported on code.google.com by [email protected]
on 27 Jul 2010 at 11:11
Purpose of code changes on this branch:
Implementing continuous sensor model
When reviewing my code changes, please focus on:
It seems like this particular section of code must be flawed in some subtle way
I'm missing, specifically symmetric headers must be ambiguous is some subtle
way.
This specifically refers to Revision #396
After the review, I'll merge this branch into
Original issue reported on code.google.com by [email protected]
on 15 Jun 2010 at 10:42
Try to build opennero according to
http://code.google.com/p/opennero/wiki/BuildingOpenNero
on a 'fresh' install of windows 7. This does not work. CMake chokes with
the following errors:
Could NOT find Doxygen (missing: DOXYGEN_EXECUTABLE)
Could NOT find ZLIB (missing: ZLIB_LIBRARIES ZLIB_INCLUDE_DIRS)
Could NOT find ZLIB (missing: ZLIB_LIBRARIES ZLIB_INCLUDE_DIRS)
Could NOT find PNG (missing: PNG_LIBRARY PNG_PNG_INCLUDE_DIR)
Could NOT find JPEG (missing: JPEG_LIBRARY JPEG_INCLUDE_DIR)
CMake Error at source/CMakeLists.txt:23 (MESSAGE):
Boost libraries were not found
I have installed the boost library, but this does not seem to make a
difference. Any thoughts on what I'm doing wrong?
Original issue reported on code.google.com by okke.formsma
on 15 Apr 2010 at 9:26
Implement a Battle mode in the NERO mod that will allow for competetive play
between trained teams.
Implement in /branches/nero then merge into trunk after completion.
Original issue reported on code.google.com by [email protected]
on 30 Aug 2010 at 2:22
as of Revision 463, the trunk version of OpenNERO fails to build on Windows.
cmake executes successfully, but bzip2 library does not build due to symbols
not found.
Original issue reported on code.google.com by ikarpov
on 26 Aug 2010 at 2:48
Steps to follow:
* Open OpenNERO/Maze
* Run and compare Q-Learning, Sarsa and rtNEAT
* Observe the fitness plots
Evaluation questions:
* Do the methods reliably solve the maze task?
* Is the performance reflected on the fitness plot?
Original issue reported on code.google.com by ikarpov
on 30 Sep 2010 at 3:36
In the Maze, the new bot needs to go through it's forward-running animation
faster so that it is in-sync with the speed at which it is transitioning in the
maze.
Original issue reported on code.google.com by ikarpov
on 22 Sep 2010 at 2:02
We need to add content to the
http://code.google.com/p/opennero/wiki/RunningOpenNeroExercise
Original issue reported on code.google.com by ikarpov
on 18 Aug 2010 at 3:46
What steps will reproduce the problem?
1. Check out the opennero code
2. Look at the directory /external/include/Python
3. It is not there
What is the expected output? What do you see instead?
I would expect to checkout all the files that are in the repository as
shown in the source code browser on this website. Not all files are
transferred.
What version of the product are you using? On what operating system?
This happens on both windows 7/tortoisesvn
TortoiseSVN 1.6.6, Build 17493 - 64 Bit , 2009/10/19 20:22:18 Subversion
1.6.6
And on fedora unix, svn version 1.4.4.
Original issue reported on code.google.com by okke.formsma
on 15 Apr 2010 at 12:54
What steps will reproduce the problem?
1. Look at the top of this screen. 'Project Home', 'Downloads' and 'Source'
are listed.
2. 'Issues' is not listed.
3. It would be nice if it were.
Can the issue tracker be enabled in the top menu?
Original issue reported on code.google.com by okke.formsma
on 15 Apr 2010 at 9:57
What steps will reproduce the problem?
Run the Roomba module a few times (either script or rtNEAT) until one of the
agents receives a reward in the very first step. If you look at the agent's
fitness in the next step, it's evident that the first step reward was doubled.
What version of the product are you using? On what operating system?
Version 403 checked out from /branches/merge/, running on Linux
Original issue reported on code.google.com by [email protected]
on 23 Jun 2010 at 12:20
bring in right-click menus from the creativeit experiment in order to allow
scaling, rotation and movement of existing objects, as well as add/remove for
walls.
Original issue reported on code.google.com by ikarpov
on 8 Jul 2010 at 7:59
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.