---
license: mit
task_categories:
- text-generation
- conversational
language:
- ja
- en
tags:
- alfworld
- agent
- trajectory
- react
- function-calling
- multi-turn
- synthetic
size_categories:
- 1K<n<10K
---

# ALFWorld Trajectory Dataset

## Overview

This is a synthetic SFT (Supervised Fine-Tuning) dataset designed for agent training in ALFWorld-compatible environments. The dataset programmatically generates expert trajectories without requiring an actual ALFWorld environment or a large language model.

### Key Approach
- **Template-based Simulation**: Lightweight simulator based on published ALFWorld information (papers, ReAct prompt examples).
- **Subgoal Decomposition**: Rule-based expert system supporting 6 task types.
- **Observation & Thought Diversity**: Various templates for diversifying observation and thought texts.
- **Multiple Formats**: Supports both function-calling and ReAct formats.

## Key Features

| Feature | Description |
|---------|-------------|
| **Compatibility** | ALFWorld-compatible tasks and actions |
| **Task Types** | 6 distinct task types: pick_and_place, clean_and_place, heat_and_place, cool_and_place, examine, pick_two |
| **Difficulty Levels** | Easy, Normal, Hard (controlling exploration detours) |
| **Trajectory Format** | Function-calling (also available in ReAct format) |
| **Multi-turn Dialogue** | Each trajectory is a multi-turn conversation between user (observation) and assistant (thought, action) |
| **Synthetic Data** | Programmatically generated, ensuring controlled and varied scenarios |

## Dataset Statistics

- **Total trajectories**: 2,502
- **Average messages per trajectory**: 24.1
- **Number of steps (turns) distribution**: 9 steps (3.9%) / 11 steps (5.0%) / 13 steps (6.7%) / 15 steps (8.4%) / 17 steps (8.4%) / 19 steps (6.8%) / 21 steps (5.9%) / 23 steps (9.9%) / 25 steps (8.0%) / 27 steps (8.0%) / 29 steps (7.8%) / 31 steps (6.1%) / 33 steps (4.7%) / 35 steps (3.4%) / 37 steps (2.3%) / 39 steps (1.8%) / 41 steps (1.2%) / 43 steps (1.0%) / 45 steps (0.5%) / 47 steps (0.1%) / 49 steps (0.1%) / 51 steps (0.0%) / 53 steps (0.1%)
- **Task type distribution**: clean_and_place: 417 (16.7%) / cool_and_place: 417 (16.7%) / examine: 417 (16.7%) / heat_and_place: 417 (16.7%) / pick_and_place: 417 (16.7%) / pick_two: 417 (16.7%)
- **Difficulty distribution**: Easy: 452 (18.1%) / Hard: 773 (30.9%) / Normal: 1277 (51.0%)

## Data Format

Each line in the JSONL file is a JSON object with the following fields:

```json
{{ # Start of outer JSON object
  "messages": [
    {{"role": "system", "content": "Interact with a household to solve a task..."}},
    {{"role": "user", "content": "You are in the middle of a room..."}},
    {{"role": "assistant", "content": "Think: I should look for apple...",
      "tool_calls": [
        {{"id": "call_1", "type": "function", "function": {{ # Inner function object
          "name": "act", "arguments": "{"action": "go to microwave 1"}"
        }}}}
      ]
    }},
    {{"role": "tool", "tool_call_id": "call_1", "content": "The microwave 1 is open..."}}
    // ... more turns
  ],
  "metadata": {{ # Metadata object
    "task_type": "heat_and_place",
    "description": "put a hot apple in/on stoveburner 1.",
    "room_type": "kitchen",
    "difficulty": "normal",
    "num_steps": 17
  }}
}}
```

## Usage

### For SFT Training
```python
from datasets import load_dataset
ds = load_dataset("u-10bei/sft_alfworld_trajectory_dataset_v4")

for sample in ds["train"]:
    messages = sample["messages"]
    # messages is already in chat format for training
```

## Last Updated
2026-02-10 04:14:33

## License
MIT