Authors: George Drayson, Efimia Panagiotaki, Daniel Omeiza, Lars Kunze

Published on: September 18, 2023

Impact Score: 8.27

Arxiv code: Arxiv:2309.09844

Summary

• What is new: Introduction of a novel approach using Heterogeneous Graph Neural Networks (HGNNs) for transforming regular driving scenarios into synthetic corner cases.
• Why this is important: Insufficiency of corner case scenarios in naturalistic driving datasets for testing autonomous vehicles.
• What the research proposes: A model that generates synthetic, realistic corner cases by perturbing the structure and properties of scene graphs representing regular driving scenarios.
• Results: Achieved 89.9% prediction accuracy in generating corner cases and demonstrated the ability to create critical situations for baseline autonomous driving methods.

Technical Details

Technological frameworks used: Heterogeneous Graph Neural Networks (HGNNs)

Models used: Attention and triple embeddings

Data used: Naturalistic driving datasets and generated scene graphs

Potential Impact

Autonomous vehicle testing and validation markets, automotive companies investing in AV development, AV safety certification agencies

Want to implement this idea in a business?

We have generated a startup concept here: SafePath AI.