File size: 2,767 Bytes
e692184
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Visual Spatial Planning (VSP) Dataset

            ## Dataset Description

            This dataset contains visual maze navigation and path verification tasks from the Visual Spatial Planning benchmark.

            ### Dataset Summary

            - **Total samples**: 1100
            - **Tasks**: Path Navigation (task-main) and Path Verification (task4)
            - **Splits**:
            - `navigation_test`: 600 samples
- `verify_test`: 500 samples

            ### Task Types

            1. **Path Verification (task4)**: Given a maze image and an action sequence, determine if the path is safe (avoids holes and reaches the goal).
            2. **Path Navigation (task-main)**: Given a maze image, generate an optimal action sequence from start to goal while avoiding holes.

            ### Data Fields

            Common fields:
            - `idx`: Unique identifier
            - `original_id`: Original data ID
            - `image`: RGB image of the maze
            - `text`: Task instruction and prompt
            - `answer`: Ground truth answer
            - `task_type`: Type of task (verify/navigation)
            - `split`: Data split (test)
            - `size`: Maze size (grid dimension)
            - `level`: Difficulty level
            - `gym_map`: Gym environment map representation (JSON string)

            Path Verification specific fields:
            - `path_length`: Length of the action sequence
            - `path`: Action sequence string

            Path Navigation specific fields:
            - `start_coords`: Starting coordinates (JSON string)
            - `goal_coords`: Goal coordinates (JSON string)
            - `obstacle_coords`: Obstacle coordinates (JSON string)
            - `astar_path`: A* algorithm path (if available)

            ### Actions

            - `L`: Move left
            - `R`: Move right
            - `U`: Move up
            - `D`: Move down

            ### Maze Elements

            - `S`: Start position
            - `G`: Goal position
            - `F`: Frozen (safe) tile
            - `H`: Hole (unsafe)

            ## Usage

            ```python
            from datasets import load_dataset

            # Load specific split
            dataset = load_dataset("hitsmy/AdaEval-VSPO", split="navigation_test")

            # Access an example
            example = dataset[0]
            image = example['image']
            prompt = example['text']
            answer = example['answer']

            ## Citation

            If you use this dataset, please cite:

            @dataset{vsp_dataset,
            title={Visual Spatial Planning Dataset},
            author={VSP Team},
            year={2024}
            }

            ##License

            Apache 2.0