Transformer Generates Diverse Mechanical Linkage Designs from Curves

The challenge of planar path synthesis involves designing mechanical linkages whose coupler curves precisely follow a specified trajectory. This problem is complex because a single curve can be generated by multiple mechanism topologies (e.g., four-, six-, or eight-bar linkages). Researchers have developed a discrete autoregressive transformer to tackle this design problem. The model is trained on a vast dataset of over one million mechanisms, using simulation-grounded evaluations to measure accuracy via Chamfer distance and dynamic time warping. The synthesis process is framed as conditional autoregressive sequence modeling, where joint coordinates are quantized into tokens and generated by a decoder-only transformer. This transformer uses a variational autoencoder (VAE) latent representation of the target curve and an explicit mechanism-type token. The approach generates diverse and accurate families of mechanisms without relying on dataset lookups, achieving impressive geometric error rates on held-out tests.