GSI

GSI Dataset (Semantic Platform): Multi-robot planning dataset with 1 scenarios, 200000 goals, 200000 tasks, and 0 prompts

Dataset Description

This dataset contains a complete multi-robot planning dataset with scenarios, goals, tasks, and prompts.

Dataset Structure

semantic/
├── tasks/              # Task definitions
│   └── {type}/
│       └── tasks.jsonl
├── scenarios/          # Scenario configurations
│   └── {type}/
│       └── {scenario_id}/
│           ├── scene_graph.json
│           ├── plans.json
│           └── scene.png
├── goals/             # Goal definitions
│   └── {type}/
│       └── goals.jsonl
└── prompts/           # Generated prompts (deduplicated)
    └── {type}/
        ├── prompts.jsonl
        ├── pool_*.json
        └── config.json

Dataset Statistics

• Types: cybertown
• Scenarios: 1
• Goals: 200000
• Tasks: 200000
• Prompts: 0

Usage

Loading the Dataset

from pathlib import Path
import json

# Load tasks
tasks_file = Path('tasks/cybertown/tasks.jsonl')
with open(tasks_file, 'r') as f:
    tasks = [json.loads(line) for line in f]

# Load scenarios
scenarios_dir = Path('scenarios/cybertown')
scenarios = {}
for scenario_dir in scenarios_dir.iterdir():
    if scenario_dir.is_dir():
        with open(scenario_dir / 'scene_graph.json', 'r') as f:
            scenarios[scenario_dir.name] = json.load(f)

# Load goals
goals_file = Path('goals/cybertown/goals.jsonl')
with open(goals_file, 'r') as f:
    goals = [json.loads(line) for line in f]

# Load prompts (deduplicated format)
prompts_dir = Path('prompts/cybertown')
# Load config
with open(prompts_dir / 'config.json', 'r') as f:
    config = json.load(f)

# Load text pools
pools = {}
for pool_file in prompts_dir.glob('pool_*.json'):
    pool_name = pool_file.stem.replace('pool_', '')
    with open(pool_file, 'r') as f:
        pools[pool_name] = json.load(f)

# Load main prompts data
with open(prompts_dir / 'prompts.jsonl', 'r') as f:
    for line in f:
        record = json.loads(line)
        # Reconstruct full prompt using indices and pools
        # record contains indices like skill_set_idx, env_desc_idx, etc.
        ...

Prompt Configuration

Citation

If you use this dataset, please cite:

@dataset{gsi,
  title = {GSI},
  author = {Windy Lab},
  year = {2025},
  license = {MIT},
}