IROS update

maps/map3.txt

@@ -0,0 +1,239 @@
+1000.00 700.00
+25 25
+0.00 700.00 0.00 0.00
+0.00 0.00 1000.00 0.00
+1000.00 0.00 1000.00 700.00
+1000.00 700.00 0.00 700.00
+210.86 0.00 210.86 167.43
+210.86 167.43 135.93 167.43
+135.93 167.43 25.14 167.43
+25.14 167.43 114.07 167.43
+114.07 167.43 114.07 0.00
+0.00 230.00 155.71 230.00
+212.86 230.00 212.86 404.29
+212.86 404.29 0.00 404.29
+274.29 0.00 274.29 60.00
+274.29 60.00 304.29 60.00
+320.00 60.00 320.00 2.86
+320.00 2.86 321.43 0.00
+341.43 60.00 371.43 60.00
+371.43 60.00 371.43 0.00
+370.00 60.00 408.57 60.00
+424.29 60.00 424.29 0.00
+440.43 60.00 470.73 60.00
+470.73 60.00 470.73 0.00
+470.73 60.00 502.05 60.00
+516.19 60.00 516.19 0.00
+533.36 60.00 562.65 60.00
+562.65 60.00 562.65 0.00
+563.67 60.00 595.06 60.00
+609.12 60.00 609.12 0.00
+274.29 80.00 274.29 153.51
+274.29 153.51 325.27 153.51
+325.27 153.51 325.27 80.00
+325.27 80.00 288.90 80.00
+325.27 80.00 364.67 80.00
+325.27 153.51 394.97 153.51
+394.97 153.51 394.97 80.00
+406.08 80.00 436.39 80.00
+436.39 80.00 436.39 153.51
+436.39 153.51 394.97 153.51
+436.39 80.00 472.75 80.00
+436.39 153.51 482.85 153.51
+482.85 153.51 482.85 80.00
+492.95 80.00 525.28 80.00
+525.28 80.00 525.28 153.51
+525.28 153.51 482.85 153.51
+525.28 80.00 547.50 80.00
+525.28 153.51 567.14 153.51
+567.14 153.51 567.14 80.00
+583.57 80.00 609.12 80.00
+609.12 80.00 609.12 153.51
+609.12 153.51 567.14 153.51
+274.29 153.51 274.29 215.72
+274.29 215.72 307.14 215.72
+325.27 215.72 325.27 153.51
+344.29 215.72 381.43 215.72
+381.43 215.72 381.43 153.51
+381.43 215.72 420.00 215.72
+437.14 215.72 437.14 153.51
+460.00 215.72 490.00 215.72
+490.00 215.72 490.00 153.51
+490.00 215.72 518.57 215.72
+531.43 215.72 531.43 153.51
+548.57 215.72 575.71 215.72
+575.71 215.72 575.71 153.51
+575.71 215.72 591.95 215.72
+609.12 215.72 609.12 153.51
+665.71 0.00 665.71 32.86
+665.71 52.86 665.71 211.43
+665.71 230.00 665.71 377.15
+665.71 377.15 1000.00 377.15
+35.36 274.73 161.62 274.73
+161.62 274.73 161.62 404.29
+336.38 263.62 289.91 263.62
+289.91 263.62 289.91 322.21
+289.91 322.21 336.38 322.21
+336.38 322.21 336.38 278.91
+389.92 323.22 389.92 262.61
+389.92 262.61 452.55 262.61
+452.55 262.61 452.55 323.22
+452.55 323.22 407.23 323.22
+493.96 277.88 493.96 318.16
+493.96 318.16 544.47 318.16
+544.47 318.16 544.47 261.60
+544.47 261.60 493.96 261.60
+356.58 370.77 356.58 438.45
+356.58 438.45 289.91 438.45
+289.91 438.45 289.91 370.77
+289.91 370.77 339.25 370.77
+574.86 405.16 574.86 361.85
+574.86 361.85 524.35 361.85
+524.35 361.85 524.35 424.47
+524.35 424.47 574.86 424.47
+410.24 416.49 410.24 377.76
+410.24 377.76 478.81 377.76
+478.81 377.76 478.81 435.22
+478.81 435.22 410.24 435.22
+0.00 473.73 337.86 473.73
+11.03 514.13 40.41 514.13
+40.41 514.13 53.88 514.13
+53.88 514.13 92.74 514.13
+92.74 514.13 119.70 514.13
+119.70 514.13 133.34 514.13
+133.34 514.13 154.02 514.13
+154.02 514.13 178.87 514.13
+178.87 514.13 207.30 514.13
+207.30 514.13 229.72 514.13
+229.72 514.13 264.15 514.13
+264.15 514.13 289.00 514.13
+289.00 514.13 321.43 514.13
+40.41 473.73 40.41 514.13
+77.78 514.13 77.78 473.73
+106.07 514.13 106.07 473.73
+133.34 514.13 133.34 473.73
+165.66 514.13 165.66 473.73
+190.92 514.13 190.92 473.73
+220.21 514.13 220.21 473.73
+244.46 514.13 244.46 473.73
+275.77 514.13 275.77 473.73
+302.04 514.13 302.04 473.73
+337.86 514.13 337.86 473.73
+0.00 542.43 29.12 542.43
+29.12 542.43 47.35 542.43
+47.35 542.43 79.58 542.43
+79.58 542.43 95.81 542.43
+95.81 542.43 128.71 542.43
+128.71 542.43 148.27 542.43
+148.27 542.43 177.16 542.43
+177.16 542.43 199.39 542.43
+199.39 542.43 235.62 542.43
+235.62 542.43 256.13 542.43
+256.13 542.43 288.91 542.43
+288.91 542.43 307.69 542.43
+307.69 542.43 330.47 542.43
+337.86 577.43 0.00 577.43
+337.86 542.43 337.86 577.43
+39.29 577.43 39.29 542.43
+65.71 577.43 65.71 542.43
+87.86 577.43 87.86 542.43
+110.71 577.43 110.71 542.43
+137.14 577.43 137.14 542.43
+160.00 577.43 160.00 542.43
+192.86 577.43 192.86 542.43
+217.86 577.43 217.86 542.43
+246.43 577.43 246.43 542.43
+274.29 577.43 274.29 542.43
+298.57 577.43 298.57 542.43
+318.57 577.43 318.57 542.43
+0.00 621.43 23.12 621.43
+23.12 621.43 44.23 621.43
+44.23 621.43 85.35 621.43
+85.35 621.43 100.46 621.43
+100.46 621.43 125.58 621.43
+125.58 621.43 140.70 621.43
+140.70 621.43 168.93 621.43
+168.93 621.43 186.04 621.43
+186.04 621.43 223.16 621.43
+223.16 621.43 244.28 621.43
+244.28 621.43 281.39 621.43
+281.39 621.43 294.51 621.43
+294.51 621.43 327.62 621.43
+327.62 621.43 337.86 621.43
+337.86 621.43 337.86 575.72
+0.00 651.43 22.65 651.43
+22.65 651.43 41.30 651.43
+41.30 651.43 85.96 651.43
+85.96 651.43 102.61 651.43
+102.61 651.43 147.26 651.43
+147.26 651.43 167.91 651.43
+167.91 651.43 210.56 651.43
+210.56 651.43 229.21 651.43
+229.21 651.43 274.52 651.43
+274.52 651.43 289.17 651.43
+289.17 651.43 349.82 651.43
+349.82 651.43 368.47 651.43
+368.47 651.43 413.12 651.43
+413.12 651.43 426.43 651.43
+426.43 651.43 426.43 700.00
+33.57 621.43 33.57 577.43
+64.29 621.43 64.29 577.43
+93.57 621.43 93.57 577.43
+132.71 621.43 132.71 577.43
+176.43 577.43 176.43 621.43
+233.57 577.43 233.57 621.43
+265.00 621.43 265.00 577.43
+287.86 621.43 287.86 577.43
+309.29 576.43 309.29 621.43
+32.86 651.43 32.86 700.00
+65.00 700.00 65.00 651.43
+94.29 651.43 94.29 700.00
+127.14 700.00 127.14 651.43
+157.86 651.43 157.86 700.00
+187.14 651.43 187.14 700.00
+220.00 700.00 220.00 651.43
+250.71 651.43 250.71 700.00
+280.71 700.00 280.71 651.43
+320.71 651.43 320.71 700.00
+360.00 700.00 360.00 651.43
+391.43 651.43 391.43 700.00
+206.43 577.43 206.43 621.43
+630.00 561.43 719.29 561.43
+719.29 561.43 719.29 497.14
+719.29 497.14 649.29 497.14
+649.29 497.14 649.29 547.86
+649.29 547.86 628.57 547.86
+652.86 561.43 652.86 595.00
+652.86 595.00 688.57 595.00
+688.57 595.00 688.57 644.29
+719.29 561.43 755.00 561.43
+755.00 561.43 755.00 644.29
+755.00 644.29 701.00 644.29
+755.00 617.15 880.00 617.15
+880.00 617.15 880.00 587.14
+755.00 570.72 880.00 570.72
+797.14 570.72 797.14 493.57
+797.14 493.57 730.71 493.57
+797.14 493.57 833.57 493.57
+833.57 493.57 833.57 523.57
+833.57 523.57 880.00 523.57
+880.00 523.57 880.00 550.00
+796.43 377.14 796.43 414.29
+796.43 414.29 842.14 414.29
+842.14 414.29 842.14 442.15
+863.57 442.15 863.57 377.14
+877.14 442.15 914.29 442.15
+914.29 442.15 914.29 407.14
+914.29 407.14 1000.00 407.14
+914.29 442.15 914.29 472.86
+914.29 472.86 987.43 472.86
+1000.00 524.29 937.14 524.29
+937.14 524.29 937.14 557.14
+937.14 568.57 937.14 608.57
+937.14 608.57 1000.00 608.57
+743.57 644.29 743.57 682.14
+820.00 617.15 820.00 700.00
+880.00 617.15 880.00 660.00
+880.00 660.00 910.00 660.00
+910.00 660.00 910.00 672.00
+910.00 672.00 910.00 700.00

src/eval_model.py

@@ -81,7 +81,7 @@ def mprint(s):
     # Eval Loop
     mprint(stamp()+"Starting model evaluation:")
     c = 0
-    for i in xrange(n_batches):
+    for i in range(n_batches):
 
         # Run training step & get next batch
         bdata = batcher.get_batch()

src/experiments/anneal_map3.py

@@ -0,0 +1,41 @@
+"""
+Copyright (C) 2017  Charles Schaff, David Yunis, Ayan Chakrabarti,
+Matthew R. Walter. See LICENSE.txt for details.
+"""
+
+# This experiment jointly learns beacon placement and localization on Map 3
+# and anneals beacon regularization throughout training.
+
+import models.inf as md
+import models.trainable_beacon as smod
+import models.prop1 as pmod
+
+md.smod = smod
+md.pmod = pmod
+
+# Number of layer blocks in the neural net
+md.nGroup = 6
+
+
+# Map file.
+MAPFILE = '../maps/map3.txt'
+
+# Determine which checkpoints to keep
+KEEPLAST = 1
+KEEPEVERY= 1e6
+SAVE_FREQ= 10000
+
+# How often to display loss compute loss under a hard beacon assignment
+DISP_FREQ=25
+HARD_FREQ=500
+
+# Optimization hyperparams
+MOM = 0.9
+WEIGHT_DECAY=0.
+BEACON_DECAY=.2
+BEACON_ANNEAL=.25
+BEACON_ANNEAL_FREQ=100000
+BSZ=1000
+
+LR = 0.01
+MAX_ITER = 1e6

src/experiments/fixed_beacon_map3.py

@@ -0,0 +1,39 @@
+"""
+Copyright (C) 2017  Charles Schaff, David Yunis, Ayan Chakrabarti,
+Matthew R. Walter. See LICENSE.txt for details.
+"""
+
+# This experiment learns to localize with our best performing fixed
+# beacon layout on Map 3.
+
+import models.inf as md
+import models.beacon14 as smod
+import models.prop1 as pmod
+
+md.smod = smod
+md.pmod = pmod
+
+# Number of layer blocks in the neural net
+md.nGroup = 6
+
+
+# Map file.
+MAPFILE = '../maps/map2.txt'
+
+# Determine which checkpoints to keep
+KEEPLAST = 1
+KEEPEVERY= 1e6
+SAVE_FREQ= 10000
+
+# How often to display loss compute loss under a hard beacon assignment
+DISP_FREQ=25
+HARD_FREQ=500
+
+# Optimization hyperparams
+MOM = 0.9
+WEIGHT_DECAY=0.
+BEACON_DECAY=0.
+BSZ=1000
+
+LR = 0.01
+MAX_ITER = 1e6

src/experiments/no_reg_map3.py

@@ -0,0 +1,38 @@
+"""
+Copyright (C) 2017  Charles Schaff, David Yunis, Ayan Chakrabarti,
+Matthew R. Walter. See LICENSE.txt for details.
+"""
+
+# This experiment jointly learns beacon placement and localization on Map 3
+# with no regularization on beacon placement.
+import models.inf as md
+import models.trainable_beacon as smod
+import models.prop1 as pmod
+
+md.smod = smod
+md.pmod = pmod
+
+# Number of layer blocks in the neural net
+md.nGroup = 6
+
+
+# Map file.
+MAPFILE = '../maps/map3.txt'
+
+# Determine which checkpoints to keep
+KEEPLAST = 1
+KEEPEVERY= 1e6
+SAVE_FREQ= 10000
+
+# How often to display loss compute loss under a hard beacon assignment
+DISP_FREQ=25
+HARD_FREQ=500
+
+# Optimization hyperparams
+MOM = 0.9
+WEIGHT_DECAY=0.
+BEACON_DECAY=0.
+BSZ=1000
+
+LR = 0.01
+MAX_ITER = 1e6

src/gen_train_data.py

@@ -54,7 +54,7 @@ def gen_train_data(mapfile, nbatch):
 
     mapfile = sys.argv[1]
     if len(sys.argv) < 3:
-        nbatch = 200
+        nbatch = 2000
     else:
         nbatch = int(sys.argv[2])
 

src/models/prop2.py

@@ -9,7 +9,7 @@
 import numpy as np
 
 WFAC   = 0.1
-DMIN   = 1.0
+DMIN   = 0.001
 
 P0     = 6.25e-4
 zeta   = 2.0

src/models/prop3.py

@@ -9,7 +9,7 @@
 import numpy as np
 
 WFAC   = 0.5
-DMIN   = 1.0
+DMIN   = 0.001
 
 P0     = 6.25e-4
 zeta   = 2.0