File size: 4,091 Bytes
2112f28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
schema_version: '1.0'
category: c
name: stacking_blocks_decomposed
description: 'The stacking task decomposed into its two primitives. T-MAT places a

  named block on the cross; T-TOP places a named block onto a pre-built stack sitting

  on the cross. Four colors: red, green, blue, yellow.'
instruction_templates:
  T-MAT: Place the {color} block on the white mat at the cross position.
  T-TOP: Place the {color} block on top of the stack at the cross position.
tasks:
  1:
    kind: T-MAT
    episodes: 60
    target: red
    instruction: Place the red block on the white mat at the cross position.
    init_stack: null
    blocks_on_black_mat: 2-4, random subset always including the target
  2:
    kind: T-MAT
    episodes: 60
    target: green
    instruction: Place the green block on the white mat at the cross position.
    init_stack: null
    blocks_on_black_mat: 2-4, random subset always including the target
  3:
    kind: T-MAT
    episodes: 60
    target: blue
    instruction: Place the blue block on the white mat at the cross position.
    init_stack: null
    blocks_on_black_mat: 2-4, random subset always including the target
  4:
    kind: T-MAT
    episodes: 60
    target: yellow
    instruction: Place the yellow block on the white mat at the cross position.
    init_stack: null
    blocks_on_black_mat: 2-4, random subset always including the target
  5:
    kind: T-TOP
    episodes: 120
    target: red
    instruction: Place the red block on top of the stack at the cross position.
    stack_heights:
    - 1
    - 2
    - 3
    episodes_per_height: 40
    height_assignment: STACK_HEIGHTS[(episode_id - 1) % 3]
    stack_colors: sampled without replacement from the three non-target colors
    distractors: 0-2, limited to the 3 - h colors left over
  6:
    kind: T-TOP
    episodes: 120
    target: green
    instruction: Place the green block on top of the stack at the cross position.
    stack_heights:
    - 1
    - 2
    - 3
    episodes_per_height: 40
    height_assignment: STACK_HEIGHTS[(episode_id - 1) % 3]
    stack_colors: sampled without replacement from the three non-target colors
    distractors: 0-2, limited to the 3 - h colors left over
  7:
    kind: T-TOP
    episodes: 120
    target: blue
    instruction: Place the blue block on top of the stack at the cross position.
    stack_heights:
    - 1
    - 2
    - 3
    episodes_per_height: 40
    height_assignment: STACK_HEIGHTS[(episode_id - 1) % 3]
    stack_colors: sampled without replacement from the three non-target colors
    distractors: 0-2, limited to the 3 - h colors left over
  8:
    kind: T-TOP
    episodes: 120
    target: yellow
    instruction: Place the yellow block on top of the stack at the cross position.
    stack_heights:
    - 1
    - 2
    - 3
    episodes_per_height: 40
    height_assignment: STACK_HEIGHTS[(episode_id - 1) % 3]
    stack_colors: sampled without replacement from the three non-target colors
    distractors: 0-2, limited to the 3 - h colors left over
prebuilt_stack:
  spawn: blocks at the cross, yaw 0, settled for 0.5 s before the episode starts
  verification: intact, upright, on the cross and at rest, else the episode is resampled
    with a derived seed
randomization:
  cross_center:
    region: centred square of side 0.1 m on the white mat
    white_mat_center:
    - 0.25
    - -0.135
    distribution: uniform
  block_xy:
    region: black mat, inset by SPAWN_MARGIN
    black_mat_center:
    - 0.25
    - 0.135
    half_extent:
    - 0.077
    - 0.067
    min_center_gap_m: 0.07
    distribution: uniform with rejection; whole set resampled if stranded
    sampled_in: canonical color order (red, green, blue, yellow)
  block_yaw_rad:
  - -0.7853981633974483
  - 0.7853981633974483
seed_formula: seed = 100_000_000*category_code + 1_000_000*task_id + episode_id +
  10_000*attempt, category_code = {a:1, b:2, c:3}; rng = numpy.random.default_rng(seed)
split_rule: episode_id % 10 == 0 -> val, else train