Scholarlink

The rapid advancement in autonomous technologies highlights the growing importance of accurate pedestrian trajectory prediction. While Generative models have demonstrated good performance in this task, their latent spaces often lack controllability. Addressing this gap, this work presents an innovative framework exploiting Generative Adversarial Networks with control vectors such as InfoGAN and augmenting them with Graph embeddings to obtain more accurate representations, which can be disentangled and improve accuracy in downstream tasks. Evaluation against benchmark datasets demonstrates good predictive accuracy for pedestrian trajectory prediction compared to current state-of-the-art approaches. Furthermore, we show that this method can enable contraollability of disentangled factors. This research lays the groundwork for developing advanced simulation tools with various real-world applications.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Leveraging Graph Networks and Generative Adversarial Networks for Controllable Trajectory Predictions

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sahib Julka,
  • Ahmad Zubair,
  • Michael Granitzer

摘要

The rapid advancement in autonomous technologies highlights the growing importance of accurate pedestrian trajectory prediction. While Generative models have demonstrated good performance in this task, their latent spaces often lack controllability. Addressing this gap, this work presents an innovative framework exploiting Generative Adversarial Networks with control vectors such as InfoGAN and augmenting them with Graph embeddings to obtain more accurate representations, which can be disentangled and improve accuracy in downstream tasks. Evaluation against benchmark datasets demonstrates good predictive accuracy for pedestrian trajectory prediction compared to current state-of-the-art approaches. Furthermore, we show that this method can enable contraollability of disentangled factors. This research lays the groundwork for developing advanced simulation tools with various real-world applications.