Reproduce “HetroD: A High-Fidelity Drone Dataset and Benchmark for Autonomous Driving in Heterogeneous Traffic” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yu-Hsiang Chen, Wei-Jer Chang, Christian Kotulla, Thomas Keutgens, Steffen Runde, Tobias Moers, Christoph Klas, Wei Zhan and 2 more

HetroD, a high-fidelity drone-view dataset, is introduced for autonomous driving in heterogeneous traffic, providing extensive annotations of vulnerable road user interactions in diverse urban settings. Evaluation of state-of-the-art models on HetroD reveals significant performance degradation in motion forecasting and planning, particularly when handling complex, unstructured multi-agent behaviors involving two-wheelers.

Key insights

  • Heterogeneous traffic, characterized by a high proportion of two-wheelers and dense, high-risk interactions, presents fundamental challenges that current state-of-the-art motion forecasting and planning models are not adequately equipped to handle.
  • Models trained on structured, vehicle-centric datasets exhibit substantial performance degradation when generalizing to heterogeneous traffic, with prediction performance highly sensitive to environmental context (map shifts).
  • Lateral VRU interactions are a significant and often overlooked failure mode for motion planners in heterogeneous environments, highlighting a need for improved interaction awareness in autonomous driving systems.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.