PCBWorld: A Benchmark Environment for Engine-Grounded PCB Design Automation

2026-07-07

AI Summary

Current learning-based approaches still struggle to match rule-based routers in printed circuit board (PCB) routing tasks that require strict adherence to design rules. This work proposes the first engine-grounded, interactive PCB routing framework, built upon the KiCad EDA engine to provide an open-source environment where agents can interact through native operations and receive real-time design rule check (DRC) feedback. The framework supports both reinforcement learning and large language models with tool-augmented reasoning, and it includes real-world and synthetic datasets alongside a unified evaluation benchmark. Experiments demonstrate that a reinforcement learning policy trained solely on synthetic data achieves zero-shot transfer to real PCBs, significantly outperforming grid-action RL and open-loop LLM baselines, and approaching the performance of conventional rule-based routers.

Abstract

PCB routing is the task of connecting the nets of a board with copper traces under strict design rules, yet learning-based methods still lag behind rule-based routers. We introduce PCBWorld, an open-source engine-grounded PCB routing environment built on the KiCad EDA engine. As a human engineer does, agents in PCBWorld interactively route a board through the engine's native operations, using its Design Rule Check (DRC) feedback to keep the routing within the design rules. The environment supports both RL policies and tool-using LLM agents. Alongside the environment, PCBWorld-Bench provides three dataset families in KiCad's native board format (.kicad_pcb), covering two types of controllable synthetic instances and 679 real open-source boards. It scores any completed board with eight engine-checked evaluation metrics, regardless of the routing method. In our experiments, agents in PCBWorld consistently outperformed grid-action RL policies and open-loop LLM baselines, and an RL policy trained only on synthetic boards transferred zero-shot to real boards, approaching rule-based routers. These results position the engine-grounded, interactive approach of PCBWorld as a promising foundation for advancing the routing ability of both RL and LLM agents.

Problem

• PCB routing
• learning-based methods
• design rule constraints
• benchmark environment
• interactive routing

Innovation

• engine-grounded routing
• interactive PCB design
• DRC-guided reinforcement learning
• tool-using LLM agents
• zero-shot transfer to real boards