Materials

Home | People | Materials | ActivitiesIranian HTM Developers | About Axaya

Axaya's Robocup Documents:
These documents provide some useful information for teams who wants to start building an agent for 3D Soccer Simulation containing Introduction to Linux, Simulation Server and Makefiles.

NOTE: Some of this documents may not respect to the new changes of 3D Soccer Simulator.

  • SPADES (Persian | English)
  • Zeitgeist Framework (Persian | English)
  • Make File (Persian | English)
  • 3D Spark Scene (Persian | English)
  • RCSS Runtime Hierarchy (Persian | English)
  • Linux Useful Commands (Persian | English)
  • Doxygen Standards (Persian | English)

All of the above documentations has been prepared for documenting the first 4 months of the team activity.

Axaya's HTM Documents:
These documents provide information about the today state of the art in HTM Technology and they would be useful for people who wants tho start Learning, Designing or Developing an HTM Network.

NOTE: Persian versions which also contains the algorithms which has been developed by Axaya will be available with some terms, until that time you can use English versions which is provided by Numenta Inc.

  • HTM Concept (Persian | English)
  • HTM Learning Algorithms (Persian | English)
  • HTM Advanced Learning Algorithms (Persian | English)
  • HTM Theory (Persian | English)
  • HTM Implementation (Persian | English)

Axaya members still trying to develop the HTM technology for applying it in Robotic applications.

Axaya's Source Code:
These source codes has been developed by Axaya Group with their own code and some adoptions which is acknowledged in the release notes and authors section of the package.
  • Basic SPADES Agent (Good for beginners)
  • HTM-Based SPAEDS Agent
  • Basic Spark Agent (Good for beginners)
  • HTM-Based Spark Agent

Source codes have some license terms that must be accepted before accessing them.

Axaya's Recognition Results:
In this section we will place our agent's log for people who wants to measure our Recognition Performance and we will welcome any interest in developing a visualizer/analyzer program for our logs.

Formate of log:
First of all the information about the initialization of the HTM network will be displayed.

Then we have sensory data processing blockes in each Simulation Step.
Examples:
====================================
[(behavior/soccer.cpp)<229>] : Sensation No. 121 Received! Processing ...
[(memory/Memory.cpp)<747>] : Flag 4 Dist Updated : 0.785079
[(memory/Memory.cpp)<784>] : Motor id : 0 Updated : 11
[(memory/Memory.cpp)<784>] : Motor id : 1 Updated : 19
[(memory/Memory.cpp)<180>] : simulateCortex():Computing FeedForward
Nodes which are processing
[(memory/Memory.cpp)<205>] : L1LN1
[(memory/Memory.cpp)<207>] : L1LN2
[(memory/Memory.cpp)<209>] : L1LN3
[(memory/Memory.cpp)<211>] : L1LN4
[(memory/Memory.cpp)<213>] : L1LN5
[(memory/Memory.cpp)<215>] : L1LN6
[(memory/Memory.cpp)<225>] : L2LN1
[(memory/Memory.cpp)<227>] : L2LN2
[(memory/Memory.cpp)<233>] : L3LN1
Input to level 1 Nodes(each node in each line)

------------------------VisionMemL1--------------------------
1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0.785079 0 0 0 0 0
0 0 0 0 0.785079 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
Output of level 1 Nodes and inputs to level 2 Nodes
------------------------VisionL1L2-----------------------------------
1-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-|-0-0-0-0-0-0-1-0-0-0-0-0-0-0-0-0-0-0-0-0-|-0-0-0-0-0-0-1-0-0-0-0-0-0-0-0-0-0-0-0-0
1-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-|-1-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-|-1-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0-0
Output of level 2 nodes and input to level 3 Nodes
------------------------VisionL2L3-----------------------------------
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Top Node Output
-----------------------VisionTopMem------------------------------
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
Real Object Stimulus: (G)for Goal Flag (F)for Corner Flag (L)for Left Side (R)for Right Side (U)for Up (D) for Down

---GLD------GLU------GRD------GRU------FLD------FLU------FRD------FRU----------------
----0--------0---------0--------0------78.5079----0--------0--------0--------0-----180
[(memory/Memory.cpp)<326>]-:-simulateCortex():Computing-FeedBack
[(memory/Memory.cpp)<426>]-:-simulateCortex():Computing-Prediction
[(memory/Memory.cpp)<522>]-:-simulateCortex():Complete
[(memory/Memory.cpp)<153>] -:-Arraies of HTM Network Resetted!

Axaya's Learning algorithm Results:
In this section we have shown the Online Learning algorithm ability in connecting the coincidences with temporal proximity, together.

Fallowing Charts shows the groups that has been discovered in a node in different levels in the soccer robot visual hierarchy.

Figure 1: The robot learned to see his own side Goal and Corner Flages.

Figure 2: In level two the Robot learned to recognize the Soccer Goal by his own (Unsupervised)
Figure 3:In the Top Node robot is able to detect the Goal and the Corner flags in every part of his visual field.

You can download all documents under our license, with AXAYA copyright.

This documents are in Persian for now and contains information that a team needs for quick start, Technical reports will be available as soon as possible.



Unpublished Documents are only available when they are published widely on the web, or only for members of
IHTMD

 

  

 

 

 

 

 

 

 

 

 Source Codes Of agent will be available for public soon, HTM-based versions will be available only for members of IHTMD






By exploring the Log files you can see the invariance abilities of our algorithms, but there would be some problems and bugs that will be solved soon.


























Temporal Pooler Charts that is the results of the new online temporal pooling mechanism implemented in Axaya Research Group.

It also supports overlapping groups, we are working on the ordered sequences and distribute representation