Decoupled Dynamics Framework with Neural Fields for 3D Spatio-temporal Prediction of Vehicle Collisions

2025-03-25

AI Summary

This work addresses the low prediction accuracy and poor generalization of existing methods for 3D vehicle collision dynamics modeling. We propose a physics-driven disentangled modeling framework featuring a novel dual-branch neural network: the Rigid Net models global rigid-body motion using quaternion representations, while the Deformation Net reconstructs local structural deformation via implicit coordinate mapping and multi-scale neural fields. Crucially, both branches are jointly optimized without component-wise supervision, enforcing physical consistency. Compared to MLP- and DeepONet-based approaches, our method achieves up to 83% lower prediction error using only 10% of the simulation data required for training. It significantly enhances generalization to extreme collision scenarios and enables high-fidelity deformation reconstruction even from low-resolution inputs.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

This study proposes a neural framework that predicts 3D vehicle collision dynamics by independently modeling global rigid-body motion and local structural deformation. Unlike approaches directly predicting absolute displacement, this method explicitly separates the vehicle's overall translation and rotation from its structural deformation. Two specialized networks form the core of the framework: a quaternion-based Rigid Net for rigid motion and a coordinate-based Deformation Net for local deformation. By independently handling fundamentally distinct physical phenomena, the proposed architecture achieves accurate predictions without requiring separate supervision for each component. The model, trained on only 10% of available simulation data, significantly outperforms baseline models, including single multi-layer perceptron (MLP) and deep operator networks (DeepONet), with prediction errors reduced by up to 83%. Extensive validation demonstrates strong generalization to collision conditions outside the training range, accurately predicting responses even under severe impacts involving extreme velocities and large impact angles. Furthermore, the framework successfully reconstructs high-resolution deformation details from low-resolution inputs without increased computational effort. Consequently, the proposed approach provides an effective, computationally efficient method for rapid and reliable assessment of vehicle safety across complex collision scenarios, substantially reducing the required simulation data and time while preserving prediction fidelity.

Problem

• Predicts 3D vehicle collision dynamics using neural fields
• Separates rigid-body motion from structural deformation explicitly
• Reduces prediction errors significantly with limited training data

Innovation

• Decoupled rigid and deformation modeling for collisions
• Quaternion-based Rigid Net for motion prediction
• Coordinate-based Deformation Net for local details