Difference between revisions of "User:Nat/Targeting Research Codename M79"

From Robowiki
Jump to navigation Jump to search
(update data)
 
(6 intermediate revisions by the same user not shown)
Line 1: Line 1:
−
{| style="background:#e8efef;border:1px solid #ccc;display:block"
+
M79 is a codename for my new targeting research. If you are around [[Eternal Rumble]] ages, please note that this research has nothing to do with [[Marshmallow]].
−
| Stage: 1 - Planned
+
 
 +
== Angular Targeting ==
 +
=== Challenges ===
 +
==== [[Targeting Challenge RM/Results|Targeting Challenge RM]] ====
 +
{| border="1" style="border-collapse: collapse; font-size: 85%; color: black"
 +
|'''Name'''||''Based on''||'''Aspd'''||'''Sprw'''||'''Fhqw'''||'''Yngw'''||'''FlMn'''||'''EASY'''||'''Tron'''||'''HTTC'''||'''RnMB'''||'''DlMc'''||'''Grbb'''||'''MEDIUM'''||'''SnDT'''||'''Cgrt'''||'''Frtn'''||'''WkOb'''||'''RkMc'''||'''HARD'''||'''TOTAL'''||'''Comments'''
 +
|-
 +
| 0001 || None || 80.71 || 86.80 || 80.76 || 92.37 || 75.30 || '''83.19''' || 79.39 || 79.19 || 79.70 || 83.03 || 69.88 || '''78.24''' || 61.76 || 67.86 || 75.50 || 75.52 || 73.66 || '''70.86''' || '''77.43''' || 36.5 seasons
 +
|-
 +
| 0002 || 0001 || 86.05 || 90.85 || 84.09 || 93.55 || 81.69 || '''87.25''' || 79.40 || 79.65 || 78.40 || 81.31 || 72.39 || '''78.23''' || 68.08 || 67.80 || 74.40 || 73.72 || 72.95 || '''71.39''' || '''78.96''' || 40.0 seasons
 +
|-
 +
| 0003 || 0002 || 84.88 || 91.18 || 84.42 || 93.46 || 82.37 || '''87.26''' || 80.88 || 79.52 || 80.53 || 82.05 || 73.53 || '''79.30''' || 67.16 || 68.66 || 74.97 || 75.55 || 74.24 || '''72.12''' || '''79.56''' || 35.0 seasons
 
|-
 
|-
−
| Current activity: Writing [[Visit Count Stats]] - [[GuessFactor Targeting (traditional)|GuessFactor Targeting]] Gun
+
|colspan="22"| '''References'''
−
|}
 
−
 
 
−
M79 is a codename for my new targeting research. If you are around [[Eternal Rumble]] ages, please note that this research has nothing to do with [[Marshmallow]]. Inspire by [[Rednaxela]]'s [[User:Rednaxela/SaphireEdge|SaphireEdge]], this is a research for a variety type of gun. Unlike [[User:Rednaxela/SaphireEdge|SaphireEdge]], each gun aimed to hit all type of robot equally, not aimed for specific type of movement.
 
−
 
 
−
== Planned Gun ==
 
−
* [[Visit Count Stats]] - [[GuessFactor Targeting (traditional)|GuessFactor Targeting]] ''' IN PROGRESS '''
 
−
* [[Pattern Matching]] - [[Play-It Forward]] - [[Multiple Choices]] (no [[Pattern Matching/Single Tick|Single Tick]])
 
−
* [[Pattern Matching]] - [[GuessFactor Targeting (traditional)|GuessFactor Targeting]] - [[Multiple Choices]]
 
−
* [[Dynamic Clustering]] - [[GuessFactor Targeting (traditional)|GuessFactor Targeting]]
 
−
* [[Dynamic Clustering]] - [[Play-It Forward]]
 
−
* [[Neural Network]] - [[GuessFactor Targeting (traditional)|GuessFactor Targeting]]
 
−
* [[Neural Network]] - [[Pattern Matching]] - [[Play-It Forward]]
 
−
 
 
−
== [[Visit Count Stats]] - [[GuessFactor Targeting (traditional)|GuessFactor Targeting]] Gun ==
 
−
{| style="background:#e8efef;border:1px solid #ccc;display:block"
 
−
| Stage: 1 - Planned
 
 
|-
 
|-
−
| Current activity: Implementing [[Visit Count Stats/Anti Aliased|Anti-Aliased Visit Count Stats]]
+
| MakoHT || || 78.30 || 79.57 || 83.60 || 95.27 || 79.94 || '''83.34''' || 80.14 || 85.46 || 75.39 || 87.01 || 55.84 || '''76.77''' || 60.37 || 66.81 || 54.13 || 72.14 || 35.42 || '''57.77''' || '''72.63''' || 43.0 seasons
 
|}
 
|}
  
−
Take a futher development path from [[Rednaxela]]'s interpolating/anti-aliasing VCS buffer. Now, planned to add an inter-bin precise angle smoothing, which allow me to use just 11 or 15 bins to saves memory but as effective as other guns with more bins. Also, I'm planning to add edges smoothing (inspire by Voidious), to make data more acculate when hit at very near GF±1. Next feature, which is my main point, is automatic segmentation. Altough I have not ever seen any robot or detailed implementation/pseudocode/algorithm discussion on this, I've several implementation detail in my head already.
+
=== Version ===
−
 
+
* '''0001''': Basic proof-of-concept. Use k-d tree to store data. Cluster size of 5; dimensions: current angle offset, velocity, distance.
−
''(I'll be really happy if [[Rednaxela]] give me his anti-aliased [[VCS]] source, but I don't think he will so I'm implementing it)''
+
* '''0002''': Cluster size of 150; exponential moving average of 50. Dimensions: current angle offset, lateral velocity, distance, acceleration, time since direction change, time since accl, wall distance forward and wall distance backward. (Fixed bug in combining factor since 0001) ''Note: seems to do much better against simple random mover, but a bit worst against more advanced mover. As expected from using more data.''
−
 
+
* '''0003''': Change angle factor to deltaAngle/MEA (should not affect score, since TC always fire at 3), weight firing wave ten times the non-firing wave.
−
==== Goals - ToDo List ====
 
−
In implementing order:
 
−
* [[Rednaxela]]'s [[Visit Count Stats/Anti Aliased|Anti-Aliased Visit Count Stats]] ''' IN PROGRESS '''
 
−
* Inter-bin precise angle smoothing
 
−
* Edges smoothing
 
−
* [[Vuen]]'s [[Automated Segmentation]] (some stuff [http://robowiki.net/cgi-bin/robowiki?AutomatedSegmentation here])
 
−
* [[Rednaxela]]'s [[Super Bot Width]]
 
−
* [[Maximum Escape Angle/Precise|Precise MEA]].
 
−
* Maybe some [[Entropy]], not sure.
 
−
* and some other I can added.
 
−
 
 
−
== Some plan to plug into robot ==
 
−
Planning to put each gun into which robot:
 
−
* First, [[VCS]] will go to [[Juno]].
 
−
* [[DC]] will go to [[Pallas]].
 
−
* All will come into one new, slow bot with [[Crowd Targeting]] :-)
 

Latest revision as of 16:37, 11 January 2011

M79 is a codename for my new targeting research. If you are around Eternal Rumble ages, please note that this research has nothing to do with Marshmallow.

Angular Targeting

Challenges

Targeting Challenge RM

Name Based on Aspd Sprw Fhqw Yngw FlMn EASY Tron HTTC RnMB DlMc Grbb MEDIUM SnDT Cgrt Frtn WkOb RkMc HARD TOTAL Comments
0001 None 80.71 86.80 80.76 92.37 75.30 83.19 79.39 79.19 79.70 83.03 69.88 78.24 61.76 67.86 75.50 75.52 73.66 70.86 77.43 36.5 seasons
0002 0001 86.05 90.85 84.09 93.55 81.69 87.25 79.40 79.65 78.40 81.31 72.39 78.23 68.08 67.80 74.40 73.72 72.95 71.39 78.96 40.0 seasons
0003 0002 84.88 91.18 84.42 93.46 82.37 87.26 80.88 79.52 80.53 82.05 73.53 79.30 67.16 68.66 74.97 75.55 74.24 72.12 79.56 35.0 seasons
References
MakoHT 78.30 79.57 83.60 95.27 79.94 83.34 80.14 85.46 75.39 87.01 55.84 76.77 60.37 66.81 54.13 72.14 35.42 57.77 72.63 43.0 seasons

Version

  • 0001: Basic proof-of-concept. Use k-d tree to store data. Cluster size of 5; dimensions: current angle offset, velocity, distance.
  • 0002: Cluster size of 150; exponential moving average of 50. Dimensions: current angle offset, lateral velocity, distance, acceleration, time since direction change, time since accl, wall distance forward and wall distance backward. (Fixed bug in combining factor since 0001) Note: seems to do much better against simple random mover, but a bit worst against more advanced mover. As expected from using more data.
  • 0003: Change angle factor to deltaAngle/MEA (should not affect score, since TC always fire at 3), weight firing wave ten times the non-firing wave.
Retrieved from "http://robowiki.net/w/index.php?title=User:Nat/Targeting_Research_Codename_M79&oldid=18066"