TC4 - ver105opt2tc4r3 - 0ebd844c-a33e-49f5-b105-8706a1e37768

[m3d]

• 2x Freyja A2700L a B2700R
• score 12 bodu

tj. zatim nase nevyssi skore, takze si skoro to issue zaslouzi ;)

[m3d]

B2700R__breadcrumb___0: 29.74 -0.32 0.00
B2700R__breadcrumb___1: 59.77 -0.34 0.00
B2700R__breadcrumb___2: 69.68 -28.45 0.00
A2700L__breadcrumb___0: 69.62 27.87 0.00
B2700R__breadcrumb___3: 69.64 -58.26 0.00
A2700L__breadcrumb___1: 42.15 39.51 0.00
B2700R__breadcrumb___4: 90.70 -0.35 0.00
B2700R__breadcrumb___5: 120.73 -0.35 0.00
A2700L__breadcrumb___2: 69.59 58.89 0.00
A2700L__breadcrumb___3: 96.88 40.36 0.00
A2700L__breadcrumb___4: 126.91 40.35 0.00
A2700L__breadcrumb___5: 149.59 59.91 0.00
min x: -6.00 y: -66.00
max x: 1576.80 y: 86.00
width: 15868px height: 1560px

[m3d]

Je to kompletni zmapovani sveta, dokonce s rezervou, kdy nektere casti videli oba roboti.

[m3d]

tj. dosah breadcrumbs je celkem maly = nedostatecny. Jinak jsou tam nejak v pravidelnem rozestupu vrtacky ;) @tajgr

[m3d]

python -m osgar.tools.lidarview --pose2d app.pose2d --lidar depth2scan.scan --keyframes detector.localized_artf --title rosmsg.sim_time_sec logs\aws\tunnel\TC4\ver105-score12\A2700L.log --rgbd detector_rear.debug_rgbd

[m3d]

A2700L ma v logu 6h simulace a 1400 sim sec:

[m3d]

Teambase z nejakeho duvodu zachytila jenom zlomek komunikace:

0:04:17.365810 received: ['TYPE_DRILL', [29749, -387, 5], 'A2700L', False]
0:06:19.612427 received: ['TYPE_DRILL', [59815, -222, 21], 'A2700L', False]
0:11:18.695973 received: ['TYPE_EXTINGUISHER', [70123, -67348, 460], 'B2700R', True]
0:11:21.413676 received: ['TYPE_DRILL', [70047, -67332, 231], 'B2700R', False]
0:12:00.277074 received: ['TYPE_DRILL', [69683, -58285, 6], 'B2700R', False]
0:14:42.916554 received: ['TYPE_DRILL', [40056, 39042, 4], 'A2700L', None]
0:14:42.917354 received: ['TYPE_DRILL', [69835, -29123, 24], 'B2700R', None]
0:16:27.972334 received: ['TYPE_DRILL', [69835, -29123, 24], 'B2700R', False]
0:16:28.216290 received: ['TYPE_DRILL', [40056, 39042, 4], 'A2700L', False]
1:01:55.051121 received: ['TYPE_DRILL', [69716, 60195, 50], 'A2700L', None]
1:01:55.051121 received: ['TYPE_RESCUE_RANDY', [151064, 83028, 337], 'A2700L', None]
1:01:55.051121 received: ['TYPE_DRILL', [149784, 61492, 28], 'A2700L', None]
1:01:55.051121 received: ['TYPE_DRILL', [128432, 40271, 73], 'A2700L', None]
1:01:55.051121 received: ['TYPE_DRILL', [124386, 40257, 35], 'A2700L', None]
1:01:55.051121 received: ['TYPE_DRILL', [98093, 40451, -27], 'A2700L', None]
1:01:55.051121 received: ['TYPE_DRILL', [69692, 28965, 42], 'A2700L', None]
1:01:55.093162 received: ['TYPE_RESCUE_RANDY', [151064, 83028, 337], 'A2700L', True]
1:01:55.093877 received: ['TYPE_DRILL', [149784, 61492, 28], 'A2700L', False]
1:01:55.873492 received: ['TYPE_DRILL', [128432, 40271, 73], 'A2700L', False]
1:01:55.874173 received: ['TYPE_DRILL', [124386, 40257, 35], 'A2700L', False]
1:01:55.874966 received: ['TYPE_DRILL', [98093, 40451, -27], 'A2700L', False]
1:01:55.875743 received: ['TYPE_DRILL', [69692, 28965, 42], 'A2700L', False]
1:01:55.876546 received: ['TYPE_DRILL', [69716, 60195, 50], 'A2700L', False]
1:07:08.136823 received: ['TYPE_DRILL', [90269, -398, -16], 'A2700L', None]

Ale jsou tam videt ty False vrtacky/breadcrumbs.

[tajgr]

Jinak jsou tam nejak v pravidelnem rozestupu vrtacky ;) @tajgr

Tak jo :-). Řešil bych to primárně doplněním trénovacích dat. Breadcrumbs tam máme jen pár, a jen z cave. Leccos by se také mohlo vyřešit naladěním thresholdů (snad již brzy).

[jisa]

Vzhledem k tomu, ze posledni trenovani mdnetu bezelo pomerne dlouho, zbyva nam cas tak na jednu iteraci uplneho pretrenovani. Tj. jestli neco potrebujeme pridat do trenovacich dat, bude potreba dat to na jeden pokus.

[m3d]

Za mne je je prijatelny hotfix na filtrovani breadcrumbs - vlastne by se mi libilo, abychom je umeli rozpoznat, ale souhlasim, ze jsou minimalne barevne stejne s vrtackou.

[jisa]

Asi se taky spis priklanim k hotfixu. Nicmene ten nemusi fungovat v pripadech, kdy robot breadcrumb po vypusteni posouva sem a tam, coz Freyja obcas dela. Pak nevime, kde ten breadcrumb vlastne je.

[m3d]

Toto info mi take prijde zajimave:

closest_artifact_report:
  name: rescue_randy_1
  type: TYPE_RESCUE_RANDY
  true_pos: 161.128 83.0188 0.39217
  reported_pos: 161.06 83.03 0.34
  distance: 0.08627775493697466
first_artifact_report: 110.448
last_artifact_report: 2930.02
mean_time_between_successful_artifact_reports: 256.3247272727273
greatest_distance_traveled: 5852.636835950343
greatest_distance_traveled_robot: B2700R
greatest_euclidean_distance_from_start: 1575.953416344078
greatest_euclidean_distance_from_start_robot: B2700R
total_distance_traveled: 11651.51510350409
greatest_max_vel: 2.411270523090993
greatest_max_vel_robot: B2700R
greatest_avg_vel: 2.281213912196342
greatest_avg_vel_robot: B2700R 

tj. mel na navrat 15min, ale jeste mu 10 min zbylo jako rezerva :) [last report] a tech 11km spolecne ujetych je take dobre :)

[m3d]

toto je ale horsi:

artifacts_found: 12
artifact_report_count: 40
duplicate_report_count: 40 

a v jinem reportu

- event:
  id: 401
  type: artifact_report_limit_reached
  time_sec: 2930
  total_score: 12 

tj. tem robotum dosly pokusy! Nelze tedy vyloucit, ze meli jeste nejake dalsi vzorky k reportovani?? (a celkove to bylo tedy "zariznuto")

[tajgr]

Vzhledem k tomu, ze posledni trenovani mdnetu bezelo pomerne dlouho, zbyva nam cas tak na jednu iteraci uplneho pretrenovani. Tj. jestli neco potrebujeme pridat do trenovacich dat, bude potreba dat to na jeden pokus.

Ale cv detektor se dá přetrénovat do dvou dnů, takže tam tich iterací se dá udělat víc.