Generates synthetic datasets for arrow-pointer chasing. The agent must traverse a grid of cells where every cell contains an arrow that dictates allowed movement directions, finding a valid step-by-step path from the start dot to a target cell — a deterministic execution of "follow the pointers" semantics.
Each sample pairs a task (first frame + prompt describing what needs to happen) with its ground truth solution (final frame showing the result + video demonstrating how to achieve it). This structure enables both model evaluation and training.
| Property | Value |
|---|---|
| Task ID | Multi-22 |
| Task | Pointer Chasing (Arrow Route) |
| Category | Algorithmic Execution |
| Resolution | 1024×1024 px |
| FPS | 16 fps |
| Duration | varies |
| Output | PNG images + MP4 video |
# 1. Clone the repository
git clone https://github.com/VBVR-DataFactory/Multi-22_pointer_chasing_arrows_data-generator.git
cd Multi-22_pointer_chasing_arrows_data-generator
# 2. Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# 3. Install dependencies
pip install --upgrade pip
pip install -r requirements.txt
pip install -e .# Generate 50 samples
python examples/generate.py --num-samples 50
# Reproducible generation with seed
python examples/generate.py --num-samples 50 --seed 42
# Custom output directory
python examples/generate.py --num-samples 100 --output data/my_dataset
# Without videos (faster, images only)
python examples/generate.py --num-samples 50 --no-videos| Argument | Description |
|---|---|
--num-samples |
Number of tasks to generate (required) |
--output |
Output directory (default: data/questions) |
--seed |
Random seed for reproducibility |
--no-videos |
Skip video generation (images only) |
[Scenario] The image displays a grid-based maze containing a red start dot, black walls, and several cells labeled with letters.
[Rules]
1. The agent must move exactly according to the provided sequence of steps.
2. The path strictly follows empty cells and does not cross walls.
[Task] Generate a video animating the agent executing the following exact moves step-by-step: right 1, up 1, up 1, up 1. The agent will eventually arrive at a target letter. Hold the final position and highlight that letter and its cell.
 |
 |
 |
| Initial Frame Grid of arrow cells + start dot + target letter |
Animation Path traced cell-by-cell along valid arrow directions |
Final Frame Complete blue path from start to target |
Trace a valid path on a grid where each cell holds an arrow that constrains the allowed movement directions out of that cell. The path must connect the start dot to the labeled target cell, drawn one step at a time as a blue line.
- Grid: 4×4 to 7×7 cells, each containing a directional arrow.
- Start: A red dot pinned to one cell.
- Target: A red letter (label) marking the destination cell.
- Movement rule: At each step, the move direction must agree with the current cell's arrow (or one of the allowed directions encoded by it).
- Goal: Draw a continuous blue trace, cell-by-cell, that reaches the target.
- Solver: BFS over the directed graph induced by the arrows guarantees a valid (shortest) path.
- Pointer-chasing semantics: Each cell encodes a local "next-step" rule, generalizing data-structure pointer traversal.
- Strict execution discipline: Skipping or violating any arrow invalidates the entire trace.
- Sequential dependency: The reachable cell at step k depends on the cells visited at steps 1..k-1.
- Visible cell-by-cell trace: Each step is rendered as a new blue segment, providing fine-grained correctness signal.
data/questions/Multi-22_pointer_chasing_arrows_data-generator_task/Multi-22_pointer_chasing_arrows_data-generator_00000000/
├── first_frame.png # Arrow grid + start + target
├── final_frame.png # Trace complete from start to target
├── prompt.txt # Task instruction
├── ground_truth.mp4 # Animation of stepwise path drawing
└── question_metadata.json # Standardized VBVR task metadata
File specifications:
- Images: 1024×1024 PNG format
- Video: MP4 format, 16 fps, H.264 + yuv420p
- Metadata: VBVR canonical schema with
task_id,vbvr_task_code,media,parameters
pointer-chasing arrow-route directed-graph algorithmic-execution deterministic path-tracing multi-step-reasoning
Part of the 36-Task Long-Horizon Multi-Step Reasoning Benchmark.