README.md

---
license: cc-by-nc-4.0
task_categories:
  - visual-question-answering
  - image-text-to-text
language:
  - en
tags:
  - embedded-systems
  - multimodal
  - procedural-reasoning
  - benchmark
  - computer-vision
pretty_name: EmbedCopilot-Bench
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*.parquet
---

# EmbedCopilot-Bench

EmbedCopilot-Bench is a multimodal benchmark for evaluating whether vision-language models can generate the next concrete instruction in embedded-device tutorials. Each example is built from consecutive tutorial action steps: the current action step provides the visual state and completed-step history, while the following action step provides the task query and ground-truth next-step instruction.

- Dataset repository: https://huggingface.co/datasets/X-EASys/EmbedCopilot-Bench
- Code and evaluation scripts: https://github.com/X-EASys/EmbedCopilot-Bench

## Dataset Structure

The released split contains 216 test examples. The split is stored as 12 Parquet shards under `data/test-*.parquet` so that the Hugging Face Dataset Viewer can load the benchmark reliably. Together, the shards form one `test` split.

Each row contains:

| Column | Type | Description |
| --- | --- | --- |
| `question_id` | string | Unique sample identifier. |
| `device_image` | image | Device-view image for the current tutorial step. |
| `screenshot_image` | image or null | Optional software/screen-view image for the current tutorial step. |
| `history_summaries` | string | Summary of tutorial steps completed before the target step. |
| `task_query` | string | User-facing query for the target next step. |
| `answer` | string | Ground-truth next-step instruction. |
| `capability` | string | Coarse capability label, for example hardware assembly or software setup. |
| `video_id` | string | Source tutorial video identifier. |
| `anchor_step_index` | int32 | Current action step index within the source tutorial. |
| `target_step_index` | int32 | Target next action step index within the source tutorial. |

## Evaluation Protocol

The benchmark prompt follows the paper's answer-generation setting: models receive the completed-step history, the visual context image or images, and the target task query, then generate the immediate method or instruction needed to accomplish the next step. The accompanying code repository provides scripts and a notebook for running model inference and judge-based evaluation with the released fields.

## Intended Use

This dataset is intended for research on multimodal procedural reasoning, embedded-device assistance, tutorial understanding, and evaluation of vision-language models on hardware/software workflows. It is designed as an evaluation benchmark rather than a training corpus.

## ESR Tasks

The code repository also includes ESR task materials for reproducibility, including task-level README files and Wokwi project files where applicable. These files document the expected executable behavior used in the evaluation setting.

## License

The dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0). The accompanying code repository includes code and scripts under the license specified in that repository.