---
license: apache-2.0
language:
  - en
pretty_name: PCEval
tags:
  - benchmark
  - physical-computing
  - arduino
  - esp32
  - circuit-design
  - code-generation
---

# PCEval Dataset

## PCEval: A Benchmark for Evaluating Physical Computing Capabilities of Large Language Models

This is the dataset accompanying the paper **“PCEval: A Benchmark for Evaluating Physical Computing Capabilities of Large Language Models,”** accepted at **NeurIPS 2026**.

**Authors:** Inpyo Song, Eunji Chon, Jangwon Lee  
**Affiliation:** Intelligence and Interactive Systems Lab, Sungkyunkwan University

[![arXiv](https://img.shields.io/badge/arXiv-2601.02404-b31b1b.svg)](https://arxiv.org/abs/2601.02404)
[![Code](https://img.shields.io/badge/GitHub-Code-181717.svg?logo=github)](https://github.com/eunjichon/PCEval-Source)
[![Project Page](https://img.shields.io/badge/Project_Page-blue)](https://songinpyo.github.io/PCEval-Project-NeurIPS2026/)

## Overview

PCEval evaluates large language models on physical computing tasks, including logical circuit generation, physical breadboard layout generation, and code generation. Each project provides a natural language description, circuit diagrams, source code, and an automated test scenario.

Projects are organized into four complexity levels:

- **Level 1:** Single component behavior verification.
- **Level 2:** Handling changing input values.
- **Level 3:** Component state changes based on inputs.
- **Level 4:** Creative projects combining multiple components.

## Project Structure

Place the downloaded project files in the code repository's `projects/` directory:

```text
PCEval-Source/
├── README.md
├── src/
├── wokwi-cli/
└── projects/
    ├── arduino/
    │   ├── level1/
    │   │   └── {project_name}/
    │   │       ├── description.md          # Project requirements
    │   │       ├── diagram.json            # Logical circuit
    │   │       ├── diagram_breadboard.json # Physical breadboard layout
    │   │       ├── scenario.yaml           # Automated test scenario
    │   │       ├── platformio.ini          # Build configuration
    │   │       ├── wokwi.toml              # Simulation configuration
    │   │       └── src/
    │   │           └── main.ino            # Project source code
    │   ├── level2/
    │   ├── level3/
    │   └── level4/
    └── esp32/
        └── ...
```

## Usage

1. **Get the code**:
   ```bash
   git clone https://github.com/eunjichon/PCEval-Source.git
   cd PCEval-Source
   ```
2. **Download the dataset** and place the project files under `projects/`, as shown above.
3. **Set up the environment** by following [How to start](https://github.com/eunjichon/PCEval-Source#how-to-start). Install PlatformIO and the modified Wokwi CLI from the [GitHub Releases](https://github.com/eunjichon/PCEval-Source/releases), add the CLI to your `PATH`, and set your `WOKWI_CLI_TOKEN`.
4. **Build and test a project**:
   ```bash
   cd projects/arduino/level1/led_blink_basic
   pio run
   wokwi-cli . --scenario scenario.yaml --diagram-file diagram.json
   # To test the physical breadboard layout:
   wokwi-cli . --scenario scenario.yaml --diagram-file diagram_breadboard.json
   ```

For evaluation scripts and experiment instructions, see the [code repository](https://github.com/eunjichon/PCEval-Source) and its [source documentation](https://github.com/eunjichon/PCEval-Source/blob/main/src/README.md).