File size: 6,616 Bytes
16405f2
e7d8e79
16405f2
 
 
 
e7d8e79
 
16405f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7d8e79
16405f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
import sys
from pathlib import Path

import pytest
import torch

REPO_ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(REPO_ROOT / "code" / "reveal_vla_bimanual"))

from models.action_decoder import ChunkDecoderConfig
from models.backbones import FrozenVLBackboneConfig
from models.multiview_fusion import MultiViewFusionConfig
from models.observation_memory import ObservationMemoryConfig
from models.planner import PlannerConfig
from models.policy import PolicyConfig
from models.reveal_head import RevealHeadConfig
from models.world_model import RevealWMConfig
from train.trainer import TrainerConfig


@pytest.fixture
def tiny_policy_config():
    def _factory(
        hidden_dim: int = 16,
        chunk_size: int = 2,
        num_candidates: int = 4,
        top_k: int = 2,
        field_size: int = 4,
        belief_map_size: int = 8,
    ) -> PolicyConfig:
        return PolicyConfig(
            backbone=FrozenVLBackboneConfig(
                hidden_dim=hidden_dim,
                freeze_backbone=True,
                gradient_checkpointing=False,
                use_dummy_backbone=True,
                depth_patch_size=8,
            ),
            fusion=MultiViewFusionConfig(
                hidden_dim=hidden_dim,
                num_layers=1,
                num_heads=4,
                ff_dim=hidden_dim * 4,
                dropout=0.0,
            ),
            memory=ObservationMemoryConfig(
                hidden_dim=hidden_dim,
                num_heads=4,
                dropout=0.0,
                history_steps=2,
                scene_history_steps=2,
                belief_history_steps=3,
                max_history_steps=4,
                scene_bank_size=2,
                belief_bank_size=2,
            ),
            decoder=ChunkDecoderConfig(
                hidden_dim=hidden_dim,
                num_heads=4,
                num_layers=1,
                ff_dim=hidden_dim * 4,
                dropout=0.0,
                chunk_size=chunk_size,
                num_candidates=num_candidates,
                num_proposal_modes=7,
                planner_top_k=top_k,
            ),
            reveal_head=RevealHeadConfig(
                hidden_dim=hidden_dim,
                num_heads=4,
                field_size=field_size,
                belief_map_size=belief_map_size,
                predict_belief_map=True,
            ),
            world_model=RevealWMConfig(
                hidden_dim=hidden_dim,
                num_heads=4,
                field_size=field_size,
                belief_map_size=belief_map_size,
                scene_bank_size=2,
                belief_bank_size=2,
            ),
            planner=PlannerConfig(
                hidden_dim=hidden_dim,
                num_heads=4,
                num_layers=1,
                num_candidates=num_candidates,
                top_k=top_k,
            ),
        )

    return _factory


@pytest.fixture
def tiny_trainer_config():
    def _factory(policy_type: str = "elastic_reveal") -> TrainerConfig:
        return TrainerConfig(
            policy_type=policy_type,
            use_bf16=False,
            gradient_checkpointing=False,
            freeze_backbone=True,
            plan_during_train=True,
            plan_during_eval=True,
        )

    return _factory


@pytest.fixture
def tiny_batch():
    def _factory(
        batch_size: int = 2,
        history_steps: int = 2,
        resolution: int = 16,
        chunk_size: int = 2,
    ) -> dict[str, torch.Tensor | list[str]]:
        images = torch.rand(batch_size, 3, 3, resolution, resolution)
        depths = torch.rand(batch_size, 3, 1, resolution, resolution)
        batch = {
            "images": images,
            "depths": depths,
            "depth_valid": torch.ones_like(depths),
            "camera_intrinsics": torch.eye(3).view(1, 1, 3, 3).expand(batch_size, 3, 3, 3).clone(),
            "camera_extrinsics": torch.eye(4).view(1, 1, 4, 4).expand(batch_size, 3, 4, 4).clone(),
            "proprio": torch.rand(batch_size, 32),
            "texts": ["test task"] * batch_size,
            "history_images": torch.rand(batch_size, history_steps, 3, 3, resolution, resolution),
            "history_depths": torch.rand(batch_size, history_steps, 3, 1, resolution, resolution),
            "history_depth_valid": torch.ones(batch_size, history_steps, 3, 1, resolution, resolution),
            "history_proprio": torch.rand(batch_size, history_steps, 32),
            "history_actions": torch.rand(batch_size, history_steps, 14),
            "action_chunk": torch.rand(batch_size, chunk_size, 14),
        }
        return batch

    return _factory


@pytest.fixture
def tiny_state():
    def _factory(batch_size: int = 2, field_size: int = 4) -> dict[str, torch.Tensor]:
        return {
            "target_belief_field": torch.rand(batch_size, 1, field_size, field_size),
            "visibility_field": torch.rand(batch_size, 1, field_size, field_size),
            "clearance_field": torch.rand(batch_size, 2, field_size, field_size),
            "occluder_contact_field": torch.rand(batch_size, 1, field_size, field_size),
            "grasp_affordance_field": torch.rand(batch_size, 1, field_size, field_size),
            "support_stability_field": torch.rand(batch_size, 1, field_size, field_size),
            "persistence_field": torch.rand(batch_size, 1, field_size, field_size),
            "reocclusion_field": torch.rand(batch_size, 1, field_size, field_size),
            "disturbance_field": torch.rand(batch_size, 1, field_size, field_size),
            "risk_field": torch.rand(batch_size, 1, field_size, field_size),
            "uncertainty_field": torch.rand(batch_size, 1, field_size, field_size),
            "access_field": torch.rand(batch_size, 3, field_size, field_size),
            "support_mode_logits": torch.rand(batch_size, 3),
            "phase_logits": torch.rand(batch_size, 5),
            "arm_role_logits": torch.rand(batch_size, 2, 4),
            "interaction_tokens": torch.rand(batch_size, 8, 16),
            "field_tokens": torch.rand(batch_size, field_size * field_size, 16),
            "latent_summary": torch.rand(batch_size, 16),
            "corridor_logits": torch.rand(batch_size, 3, 32),
            "persistence_horizon": torch.rand(batch_size, 3),
            "disturbance_cost": torch.rand(batch_size),
            "reocclusion_logit": torch.rand(batch_size, 3),
            "belief_map": torch.rand(batch_size, 1, 8, 8),
            "compact_state": torch.rand(batch_size, 30),
        }

    return _factory