MechaFormer: Sequence Learning for Kinematic Mechanism Design Automation

2025-08-12

AI Summary

Mechanism path-following design faces a high-dimensional, complex search challenge arising from the coupling of topological discreteness and parametric continuity. Method: This paper introduces the first Transformer-based end-to-end generative framework for mechanism synthesis. It defines a domain-specific language (DSL) tailored to mechanism design, enabling unified encoding of both topology and continuous parameters; the path-to-mechanism mapping is formulated as a conditional sequence generation task. The framework supports diverse sampling and integrates seamlessly with conventional optimizers to form an efficient hybrid solving strategy. Contribution/Results: Experiments demonstrate state-of-the-art trajectory matching accuracy, substantially improved search efficiency, generation of numerous novel and kinematically feasible mechanisms, and provision of high-quality initial solutions for downstream optimization. This work establishes, for the first time, a learnable and generative paradigm for mechanism design.

Technology Category

Search and Optimization: Sampling/Simulation-based SearchMultiagent Systems: Mechanism DesignIntelligent Robots: Motion and Path Planning

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingResponsible Web: Machine-in-the-loop, human agency and autonomyEconomics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Designing mechanical mechanisms to trace specific paths is a classic yet notoriously difficult engineering problem, characterized by a vast and complex search space of discrete topologies and continuous parameters. We introduce MechaFormer, a Transformer-based model that tackles this challenge by treating mechanism design as a conditional sequence generation task. Our model learns to translate a target curve into a domain-specific language (DSL) string, simultaneously determining the mechanism's topology and geometric parameters in a single, unified process. MechaFormer significantly outperforms existing baselines, achieving state-of-the-art path-matching accuracy and generating a wide diversity of novel and valid designs. We demonstrate a suite of sampling strategies that can dramatically improve solution quality and offer designers valuable flexibility. Furthermore, we show that the high-quality outputs from MechaFormer serve as excellent starting points for traditional optimizers, creating a hybrid approach that finds superior solutions with remarkable efficiency.

Problem

• Automates kinematic mechanism design for path tracing
• Integrates topology and parameter optimization in one process
• Enhances solution quality with hybrid optimization approach

Innovation

• Transformer-based model for mechanism design
• Conditional sequence generation for topology and parameters
• Hybrid approach combining learning and traditional optimization