AI RESEARCH
SD2AIL: Adversarial Imitation Learning from Synthetic Demonstrations via Diffusion Models
arXiv CS.LG
•
ArXi:2512.18583v2 Announce Type: replace Adversarial Imitation Learning (AIL) is a dominant framework in imitation learning that infers rewards from expert nstrations to guide policy optimization. Although providing expert nstrations typically leads to improved performance and greater stability, collecting such nstrations can be challenging in certain scenarios. Inspired by the success of diffusion models in data generation, we propose SD2AIL, which utilizes synthetic nstrations via diffusion models.