A Deep Learning Framework for Generation and Analysis of Driving Scenario Trajectories

Abstract

We propose a unified deep learning framework for the generation and analysis of driving scenario trajectories, and validate its effectiveness in a principled way. To model and generate scenarios of trajectories with different lengths, we develop two approaches. First, we adapt the Recurrent Conditional Generative Adversarial Networks (RC-GAN) by conditioning on the length of the trajectories. This provides us the flexibility to generate variable-length driving trajectories, a desirable feature for scenario test case generation in the verification of autonomous driving. Second, we develop an architecture based on Recurrent Autoencoder with GANs to obviate the variable length issue, wherein we train a GAN to learn/generate the latent representations of original trajectories. In this approach, we train an integrated feed-forward neural network to estimate the length of the trajectories to be able to bring them back from the latent space representation. In addition to trajectory generation, we employ the trained autoencoder as a feature extractor, for the purpose of clustering and anomaly detection, to obtain further insights into the collected scenario dataset. We experimentally investigate the performance of the proposed framework on real-world scenario trajectories obtained from in-field data collection.

Scenario-Based Trajectory Generation and Density Estimation Towards Risk Analysis of Autonomous Vehicles

“…While this compares between models well, it does not give a concrete value of how well the models perform. Because of this, we used two metrics, matching and coverage, which were introduced in [10]. Matching measures how "realistic", in terms of how similar the generated trajectories are to the real data.…”

“…G m is the set of m generated trajectories, R n the set of n real trajectories, and dist is some function defining the distance between two trajectories. In our case, as in [10], we used Dynamic Time Warping (DTW) [17].…”

“…TrajVAE and TrajGAN introduced by [9] use Long Short Term Memory (LSTM) cells and an embedding to lower the dimensionality. Lastly, in the work by [10], two types of GANs are adapted for trajectory generation. In each of these papers, a different method has been used in order to adapt generative models for trajectory generation.…”

Research on Specific Scenario Generation Methods for Autonomous Driving Simulation Tests

“…The US NGSIM dataset, which uses high-definition cameras set up next to highways to collect highway vehicle data, has collected more than 10,000 pieces of data for the study of high-speed tailgating behavior. Accident scene datasets are generally used for the generation of hazard scenes, and Demetriou et al [10] used the RC-GAN method to establish a deep learning framework based on the driving trajectory data in real accident data to realize the extension of vehicle driving trajectories so as to generate a hazard scene library. Ding et al [11] proposed a multi-vehicle trajectory generator consisting of a bidirectional encoder and a multi-branch decoder.…”

The Learning Trajectory Based on STEM of Elementary School Pupils’ in Solving Proportion Material: Didactical Design-Research

“…Eudoxus developed the theory of ratios in the 4th century BCE, which is still used today to describe the relationship between two ratios and their values (Demetriou, 2023). Al-Khwarizmi, a mathematician and astronomer from the 9th century, was a famous figure in the development of proportionality in mathematics (Tykhonov, 2023).…”