Discrete Autoregressive Transformer for Generative Mechanism Synthesis

The article presents a Discrete Autoregressive Transformer for synthesizing mechanisms that match specified trajectories, utilizing a dataset of over one million mechanisms. The model employs a decoder-only transformer architecture, integrating a variational autoencoder (VAE) for latent representation and conditional autoregressive sequence modeling, achieving a mean Chamfer distance of 0.0132 and mean dynamic time warping of 0.153 on held-out tests. This approach enables diverse and accurate mechanism generation without the need for dataset lookup, which is significant for practitioners in robotics and mechanical design seeking efficient synthesis of complex mechanisms.